On topic: What a surprise. Rhetoric isnt action. Maybe a look at job vacancy data and employer intention will give you an idea of the possible damage the budget may have done to sentiment and hard financial forecasts by companies.
Off topic. Israel is reported to have launched about 150 'airstrikes' on targets in Syria. We don't know what that 150 relates to; launches, separate targets or sorties but any way you look at it, its a lot. Notably strikes have occurred at the old Russian naval facility at Latakkia. The indictions are that the current Syrian government is open to allowing Russia to retain a presence but Israel might conclude that it needs to lob some more at those former facilities so they arent easy to walk into. Ostensibly these strikes are designed to take out chemical weapon related facilities and heavy missiles but, they are also attacking the likes of air defence capabilities and storage sites. This suggests part of the plan is to support Israels ability to act with impunity in Syria if it needs to.
Meanwhile a number of hundred Russian troops and military contractors are a bit stuck at the moment. Russia has asked the Turks to help out and the likelihood is they will get out ok, but its a worry given some serious local hostility.
If you ask me - going just off my gut reaction - it's surely at least a 70% chance he doesn't plead guilty by June. What possible reason would have to do so? Only the plea bargain of the century and it seems unlikely he's offered that. Even if he wants to, does he have the opportunity to do it by then? If I were him I'd be doing my best to get an OJ calibre jury.
It's worth remembering that juries in NY State trials work on unanimous decisions, not majorities, iirc. That effects the likelihood of pleas.
When Chump was up for his E Jean Carroll charges, it was all about identifying a single juror who followed only Trumpish news sources, in the hope that there would be one nail-Juror.
If you ask me - going just off my gut reaction - it's surely at least a 70% chance he doesn't plead guilty by June. What possible reason would have to do so? Only the plea bargain of the century and it seems unlikely he's offered that. Even if he wants to, does he have the opportunity to do it by then? If I were him I'd be doing my best to get an OJ calibre jury.
His best bet will be a mental health excuse. He left social media abruptly some months ago, so the foundation for a breakdown narrative is there. So, not guilty by reason of mental disorder or defect.
The rumoured spine operation and extreme chronic pain will be the strategy. But there is a chance he wants to go down as a martyr.
Martyrs don't obtain illegal firearms and kill in cold blood.
rcs1000 said: "CCS is one of those things where the more you know about it, the more you know how ridiculous it is."
Please don't tell the trees this; they might go on strike.
(Freeman Dyson has some interesting thoughts on this problem,)
Trees are great. But they are also very temporary stores. At the forest level they can act as long term storage, but it just takes one fire to release all that CO2 back into the atmosphere.
Personally, I think we should just relax about the use of gas. It'll get eventually displaced by solar, but in the interim it's better than most of the alternatives.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
Also, a reminder of just fori good the open source LLMs are getting.
He could do with using it to update the look of his website...Less 2024, more 1994.
It's just a blog post about using Ollama. But it is incredible how good the open source models are, and how wrong people are who think that a few AI companies are going to monopolize the space.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
Kudos to Israel, sounds like a very smart move that will hopefully lead to improved security for them there. 👍
You approve of taking advantage of instability in a neighbouring country to seize strategic territory? How about Putin and Crimea?
One of those occasions when I agree with you, william.
There's no good justification for seizing territory in this manner. That way international chaos lies.
You can't approve it just because it's an ally.
Of course you can.
They're at war. Why can't states at war do their best to make themselves secure?
You and William are pretending Syria is some poor, defenceless, peaceful state that is being attacked unprovoked like Ukraine was, rather than a party to an ongoing war.
Syria is not at war with anyone right now.
There might be justification for a temporary buffer zone given the prevailing uncertainty - but a land grab is simply illegal.
Zero debate about that.
Syria is at war with Israel right now.
Should have signed a peace treaty while they had the chance.
They've only had 76 years of war to do so.
Amazed Nigel was ignorant of the fact that Israel and Syria are at war.
Right now that simply isn't the case. The Syrian regime has been overthrown.
If Israel tries to permanently seize territory, then it's quits likely we'll see another war some time in the next few years.
If Israel don't want be part the western liberal democracies, and just another set of Middle East warlords, that will have consequences.
Right now it simply is the case.
Israel has been at war with Syria for 76 years and counting, predating the Assad regime.
You pretend the war is not real, but for Israel it absolutely is. If the new Syrian government wishes to sue for peace, then great, but if it does not then Israel is well within her rights to persecute the war within the confines of the law and that doesn't prevent taking action against states you are currently at war with.
I'm not pretending anything. I am saying that the permanent seizure of territory is completely unjustified.
It's very pretty clear from Israeli cabinet statement thaf is what is intended.
On topic: What a surprise. Rhetoric isnt action. Maybe a look at job vacancy data and employer intention will give you an idea of the possible damage the budget may have done to sentiment and hard financial forecasts by companies.
Off topic. Israel is reported to have launched about 150 'airstrikes' on targets in Syria. We don't know what that 150 relates to; launches, separate targets or sorties but any way you look at it, its a lot. Notably strikes have occurred at the old Russian naval facility at Latakkia. The indictions are that the current Syrian government is open to allowing Russia to retain a presence but Israel might conclude that it needs to lob some more at those former facilities so they arent easy to walk into. Ostensibly these strikes are designed to take out chemical weapon related facilities and heavy missiles but, they are also attacking the likes of air defence capabilities and storage sites. This suggests part of the plan is to support Israels ability to act with impunity in Syria if it needs to.
Meanwhile a number of hundred Russian troops and military contractors are a bit stuck at the moment. Russia has asked the Turks to help out and the likelihood is they will get out ok, but its a worry given some serious local hostility.
Israel (somewhat inadvertently) won a huge strategic victory by precipitating the Syrian revolution. It potentially cuts off both Iranian and Russian influence in Syria, and the weapons supplies that have been used against them.
By immediately attacking the new regime, they risk what might have been an end to active hostilities with Syria. And potentially undoing that gain.
If you ask me - going just off my gut reaction - it's surely at least a 70% chance he doesn't plead guilty by June. What possible reason would have to do so? Only the plea bargain of the century and it seems unlikely he's offered that. Even if he wants to, does he have the opportunity to do it by then? If I were him I'd be doing my best to get an OJ calibre jury.
His best bet will be a mental health excuse. He left social media abruptly some months ago, so the foundation for a breakdown narrative is there. So, not guilty by reason of mental disorder or defect.
The rumoured spine operation and extreme chronic pain will be the strategy. But there is a chance he wants to go down as a martyr.
Yup. He needs strong lawyers to tell him what is best. Coming from a rich family, I suspect he will get them.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
That isn't how most successful social media companies go though, they pretty much all hockey stick due to network effect and fast once awareness is raised about their existence and they go through phase of doubling every year for 4-5 years until they hit their own saturation level.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
That isn't how most successful social media companies go though, they pretty much all hockey stick due to network effect and fast once awareness is raised about their existence and they go through phase of doubling every year for 4-5 years until they hit their own saturation level.
I think that's true if - and only if - you are filling a need that was previously unfulfilled. In the case of Bluesky, it is offering a near identical service to Twitter, only without Elon.
Now, either Bluesky is going forward or it's going backward. But right now, it is very definitely going forward.
I also suspect that it will never end up with a 100% share, like Twitter used to have. The internet is fragmenting, and people are choosing to only see things they agree with. Bluesky will therefore never hockeystick to 100% of the "twitter"-type space, because it only appeals to the (substantial) minority of people who want Twitter and hate Elon.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I think "stolen" is not entirely accurate. Yes: a lot of the training data was "purloined".
But there's an awful lot of open source data out there, including much of the open web. If you scrape the whole of public GitHub, you will know almost all you could ever need to know about coding.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
That isn't how most successful social media companies go though, they pretty much all hockey stick due to network effect and fast once awareness is raised about their existence and they go through phase of doubling every year for 4-5 years until they hit their own saturation level.
I disagree with that. *Early* social media companies went hockey-stick, because they were operating in a near-vacuum. There were hundreds of millions of people on t'Internet, and no-one had grabbed the Twitter-style market. Yes, Twitter had rivals, but they were also small, or small parts of a massive organisation that the organisation did not care for much.
Now, there are incumbents. Bluesky is not operating in a vacuum; it is an insurgent against a massive competitor. And that's far more difficult than trying to build something where, at worst, your competitors are the same size as you.
It is much harder to go hockey-stick against a massive incumbent. The fact Bluesky has taken off as it has is either a sign they're brilliant, and/or that Twix is mucking up badly.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
That isn't how most successful social media companies go though, they pretty much all hockey stick due to network effect and fast once awareness is raised about their existence and they go through phase of doubling every year for 4-5 years until they hit their own saturation level.
I disagree with that. *Early* social media companies went hockey-stick, because they were operating in a near-vacuum. There were hundreds of millions of people on t'Internet, and no-one had grabbed the Twitter-style market. Yes, Twitter had rivals, but they were also small, or small parts of a massive organisation that the organisation did not care for much.
Now, there are incumbents. Bluesky is not operating in a vacuum; it is an insurgent against a massive competitor. And that's far more difficult than trying to build something where, at worst, your competitors are the same size as you.
It is much harder to go hockey-stick against a massive incumbent. The fact Bluesky has taken off as it has is either a sign they're brilliant, and/or that Twix is mucking up badly.
Tiktok went hockey stick against all the established players. As did Telegram.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I think "stolen" is not entirely accurate. Yes: a lot of the training data was "purloined".
But there's an awful lot of open source data out there, including much of the open web. If you scrape the whole of public GitHub, you will know almost all you could ever need to know about coding.
I think stolen sums it up, especially for image and coding data.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I think this is true, although it is worth pointing out that Llama isn't produced by some random nobodies, its the might of Meta, it is just Zuckerberg has decided on the path of open sourcing it.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
That isn't how most successful social media companies go though, they pretty much all hockey stick due to network effect and fast once awareness is raised about their existence and they go through phase of doubling every year for 4-5 years until they hit their own saturation level.
I disagree with that. *Early* social media companies went hockey-stick, because they were operating in a near-vacuum. There were hundreds of millions of people on t'Internet, and no-one had grabbed the Twitter-style market. Yes, Twitter had rivals, but they were also small, or small parts of a massive organisation that the organisation did not care for much.
Now, there are incumbents. Bluesky is not operating in a vacuum; it is an insurgent against a massive competitor. And that's far more difficult than trying to build something where, at worst, your competitors are the same size as you.
It is much harder to go hockey-stick against a massive incumbent. The fact Bluesky has taken off as it has is either a sign they're brilliant, and/or that Twix is mucking up badly.
Tiktok went hockey stick against all the established players.
Were those players as established in China, where TikTok took off? (I'd also argue that TikTok's service was rather different to the 'established' players back in 2016/7).
Also, they were not small. Less than a year after launching, they spent a billion dollars buying a Yankian firm. They had very big Chinese backers.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I think "stolen" is not entirely accurate. Yes: a lot of the training data was "purloined".
But there's an awful lot of open source data out there, including much of the open web. If you scrape the whole of public GitHub, you will know almost all you could ever need to know about coding.
I think stolen sums it up, especially for image and coding data.
Errr:
If your code is open source and you dumped it in Github, then that's not stolen.
You chose to make it available under an open source license, and I don't really see why people shouldn't be allowed to use it to train their model.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
That isn't how most successful social media companies go though, they pretty much all hockey stick due to network effect and fast once awareness is raised about their existence and they go through phase of doubling every year for 4-5 years until they hit their own saturation level.
I disagree with that. *Early* social media companies went hockey-stick, because they were operating in a near-vacuum. There were hundreds of millions of people on t'Internet, and no-one had grabbed the Twitter-style market. Yes, Twitter had rivals, but they were also small, or small parts of a massive organisation that the organisation did not care for much.
Now, there are incumbents. Bluesky is not operating in a vacuum; it is an insurgent against a massive competitor. And that's far more difficult than trying to build something where, at worst, your competitors are the same size as you.
It is much harder to go hockey-stick against a massive incumbent. The fact Bluesky has taken off as it has is either a sign they're brilliant, and/or that Twix is mucking up badly.
Tiktok went hockey stick against all the established players. As did Telegram.
Because it offered a 60 second video platform (and just that and no people sharing photos of their cat), and no one else did. And it combined this with a fantastic recommendation algorithm that showed you things you wanted to watch.
But this is all irrelevant.
There's more than one way to skin a cat. Some services will go hyperbolic. Others will be slower constant growers.
And most - of course - will fail.
There's no immutable law that says that if you're growing at 10% a month rather than 25%, then you're doomed.
Look at mobile gaming, there are plenty of examples of games which went 1, 100, 10,000, and then all the way back to 1 very quickly. While other games, have grown much more slowly, but have shown much greater sticking power.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
That isn't how most successful social media companies go though, they pretty much all hockey stick due to network effect and fast once awareness is raised about their existence and they go through phase of doubling every year for 4-5 years until they hit their own saturation level.
I think that's true if - and only if - you are filling a need that was previously unfulfilled. In the case of Bluesky, it is offering a near identical service to Twitter, only without Elon.
Now, either Bluesky is going forward or it's going backward. But right now, it is very definitely going forward.
I also suspect that it will never end up with a 100% share, like Twitter used to have. The internet is fragmenting, and people are choosing to only see things they agree with. Bluesky will therefore never hockeystick to 100% of the "twitter"-type space, because it only appeals to the (substantial) minority of people who want Twitter and hate Elon.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I think "stolen" is not entirely accurate. Yes: a lot of the training data was "purloined".
But there's an awful lot of open source data out there, including much of the open web. If you scrape the whole of public GitHub, you will know almost all you could ever need to know about coding.
I think stolen sums it up, especially for image and coding data.
Errr:
If your code is open source and you dumped it in Github, then that's not stolen.
You chose to make it available under an open source license, and I don't really see why people shouldn't be allowed to use it to train their model.
I disagree, especially when the entity stealing the data makes money from it. And it's even more egregious with image-sharing sites.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
Yes, it was the second story on the main evening news after a poll about the Treaty Principles Bill which has been proposed by New Zealand’s answer to Nigel Farage, David Seymour, leader of ACT.
The Racing Minister, Winston Peters, played the “it’s all about the welfare of the dogs” card and it’s probably supported by the majority over here. More than 1,000 people will lose their jobs but the tracks are candidates for redevelopment and that will bring in some money to soften the blow.
The last dog meeting will likely be in July 2026. That will leave only the UK, Ireland, Australia and parts of the USA. Apparently it’s legal in Mexico and Vietnam but there are no tracks - as to whether there are any dogs, I’ll leave that to others.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
It has been steadily climbing though: as recently as three months ago, i was seeing a ratio of 100-1 Twitter to Blue sky posts. Now it's maybe 4-1. And every day it gets a little more even.
Yes it is, but about 3-4 weeks ago we had the big media hype and that kicked up the users numbers a fair bit but it then didn't kick off the hockey stick, more gradual ticking up.
Bluesky doesn't need hockey stick, it just needs to keep taking tiny bit of market share, by tiny bit of market share.
That isn't how most successful social media companies go though, they pretty much all hockey stick due to network effect and fast once awareness is raised about their existence and they go through phase of doubling every year for 4-5 years until they hit their own saturation level.
I disagree with that. *Early* social media companies went hockey-stick, because they were operating in a near-vacuum. There were hundreds of millions of people on t'Internet, and no-one had grabbed the Twitter-style market. Yes, Twitter had rivals, but they were also small, or small parts of a massive organisation that the organisation did not care for much.
Now, there are incumbents. Bluesky is not operating in a vacuum; it is an insurgent against a massive competitor. And that's far more difficult than trying to build something where, at worst, your competitors are the same size as you.
It is much harder to go hockey-stick against a massive incumbent. The fact Bluesky has taken off as it has is either a sign they're brilliant, and/or that Twix is mucking up badly.
Tiktok went hockey stick against all the established players. As did Telegram.
Because it offered a 60 second video platform (and just that and no people sharing photos of their cat), and no one else did. And it combined this with a fantastic recommendation algorithm that showed you things you wanted to watch.
But this is all irrelevant.
There's more than one way to skin a cat. Some services will go hyperbolic. Others will be slower constant growers.
And most - of course - will fail.
There's no immutable law that says that if you're growing at 10% a month rather than 25%, then you're doomed.
Look at mobile gaming, there are plenty of examples of games which went 1, 100, 10,000, and then all the way back to 1 very quickly. While other games, have grown much more slowly, but have shown much greater sticking power.
This isn't actually true. Originally it was a Vine ripoff, Vine was 7s vs 12s Tiktok if I remember correctly. The discontinuation of Vine came as many different competing platforms began to introduce their own equivalents to Vine's short-form video approach and the thought was there was no money in this because who wants to watch a 30s ad between 10s vids and anybody can spin up such a site. The secret to TikTok success, the recommendation algorithm, something that even YouTube doesn't match today, and it went hockey stick in China and again in the West because of it.
I have always said, even before Musk, twitter as a product is crap and they crucially (unlike Meta) don't know their customers, which always makes selling ads harder and less valuable. Before Musk, there was virtually no feature development.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I think "stolen" is not entirely accurate. Yes: a lot of the training data was "purloined".
But there's an awful lot of open source data out there, including much of the open web. If you scrape the whole of public GitHub, you will know almost all you could ever need to know about coding.
I think stolen sums it up, especially for image and coding data.
Errr:
If your code is open source and you dumped it in Github, then that's not stolen.
You chose to make it available under an open source license, and I don't really see why people shouldn't be allowed to use it to train their model.
I disagree, especially when the entity stealing the data makes money from it. And it's even more egregious with image-sharing sites.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
I think "stolen" is not entirely accurate. Yes: a lot of the training data was "purloined".
But there's an awful lot of open source data out there, including much of the open web. If you scrape the whole of public GitHub, you will know almost all you could ever need to know about coding.
I think stolen sums it up, especially for image and coding data.
Errr:
If your code is open source and you dumped it in Github, then that's not stolen.
You chose to make it available under an open source license, and I don't really see why people shouldn't be allowed to use it to train their model.
I disagree, especially when the entity stealing the data makes money from it. And it's even more egregious with image-sharing sites.
Unsurprising. The Israelis have been nicking other people's land for a century. The weird thing is how they then act surprised when everybody else is furious. They never seem to learn. Of course they'll kick a neighbour who is down.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
However, random people in their back bedroom can't train LLMs and even with a few nVidia 9090s in 2040 you won't be able to. We aren't going to get from needing 100k GPUs and billions a year on compute costs to home usage in the foreseeable future. So if they put in place a whole amount of regulation, you won't be able to train SOTA models nor will small startups and they will lag behind the authorised ones which will be a small number of companies.
Now could some people train some small models on something specialist yes, and open source them, but its a bit like you can hack games consoles, people do, can you buy Raspberry Pi and do emulation, sure, but Sony, Microsoft and Nintendo are the only ones making them for the masses. And as we see with emulation, the big companies absolutely monster these people making open source emulation software and tie them up time and time again with legal attacks.
The obvious way this gets shut out to the masses is that huge chunks of the training data is protected, its theft to use it, and only these small number of regulated companies are allowed to use it...the fact they got there by "acquiring" it will be seen as simply them innovating to start with and now they are very sorry and they make sure they make big payments for the right to use it.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
Oh indeed, but politicians have no understanding and lobbyists know how to push their buttons. Something as simple as a requirement for $100m of liability insurance is probably all that’s needed.
Talking of banning maths, the recent quantum computing developments are, if replicable, potentially even more world-changing.
Jury "nullification" must be a risk for the prosecution of the CEO killer. Just astonishing levels of support, with the McDonald's where he was apparently reported to the police swamped with negative reviews (it was a "boomer patron" according to the internet).
One of the most interesting things to come out of the US for some time, particularly in the context of Trump's election. People keep on talking about wealth inequality being higher in the US now than it was at the start of the French Revolution...
Very disturbing that so many people appear to support the worse crime that anyone can commit.
There’s no evidence of pineapple on pizza, Radiohead or coding in Python.
Unsurprising. The Israelis have been nicking other people's land for a century. The weird thing is how they then act surprised when everybody else is furious. They never seem to learn. Of course they'll kick a neighbour who is down.
Yes and no. Most of the Golan Heights were grabbed during the Six-Day war, which was one of a series of wars - and ones where the Arabs wanted to grab Israeli territory. In fact, in some of the wars they wanted to destroy Israel.
Also Israel has been willing to hand land back in the past: the Sinai being an extremely large example.
So it is more complex than that. But on the other hand, the settlers...
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
Oh indeed, but politicians have no understanding and lobbyists know how to push their buttons. Something as simple as a requirement for $100m of liability insurance is probably all that’s needed.
Talking of banning maths, the recent quantum computing developments are, if replicable, potentially even more world-changing.
Unsurprising. The Israelis have been nicking other people's land for a century. The weird thing is how they then act surprised when everybody else is furious. They never seem to learn. Of course they'll kick a neighbour who is down.
Yes and no. Most of the Golan Heights were grabbed during the Six-Day war, which was one of a series of wars - and ones where the Arabs wanted to grab Israeli territory. In fact, in some of the wars they wanted to destroy Israel.
Also Israel has been willing to hand land back in the past: the Sinai being an extremely large example.
So it is more complex than that. But on the other hand, the settlers...
This is the one major part of Israels behaviour that you don't need to throw your load in with the anti-Semitic hate mob that want to "Free Palestine from the river to the sea" to comfortably state is absolutely wrong.
Jury "nullification" must be a risk for the prosecution of the CEO killer. Just astonishing levels of support, with the McDonald's where he was apparently reported to the police swamped with negative reviews (it was a "boomer patron" according to the internet).
One of the most interesting things to come out of the US for some time, particularly in the context of Trump's election. People keep on talking about wealth inequality being higher in the US now than it was at the start of the French Revolution...
Very disturbing that so many people appear to support the worse crime that anyone can commit.
I'd argue that it was far from the (worst?) crime anyone can commit. It was cold-blooded murder, but at least it was quick and efficient.
How does that compare to (say) a father raping a child? Or a decades-long abusive relationship between husband and wife? Or the stories of torture and murder we have seen coming out of Sednaya Prison? Or a gunman running amock in a school, killing at random?
The guy did an evil, but it is by far the deepest evil.
(I am not defending what he did; he should go to jail for life.)
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
Oh indeed, but politicians have no understanding and lobbyists know how to push their buttons. Something as simple as a requirement for $100m of liability insurance is probably all that’s needed.
Talking of banning maths, the recent quantum computing developments are, if replicable, potentially even more world-changing.
F1: won't be for a little while but when I put up a post-season betting review the graph is going to be rather lopsided. I was behind almost the entire season until the last half dozen races, all but one of which was green.
F1: won't be for a little while but when I put up a post-season betting review the graph is going to be rather lopsided. I was behind almost the entire season until the last half dozen races, all but one of which was green.
F1 is on today!!
Post-season test taking place in Abu Dhabi, with each team using one of the regular drivers for tyre testing, and a young driver.
As soon as we can totally choke them of hard currency transfers for O&G sales to China and India.
Interest rates are 20% as well, and the rouble lost 25% against the USD between August and last week, since when it’s bounced a little.
The ruble may, or may not, have bounced. The Russian central bank has paused foreign currency purchases until the end of the year, which makes the value of the ruble hard to discern. It is also allegedly selling the Yuan to prop the ruble up. AIUI, it's value is pretty artificial atm.
(Though again AIUI, they did the same thing last year, and the ruble's value recovered when trading resumed.)
Unsurprising. The Israelis have been nicking other people's land for a century. The weird thing is how they then act surprised when everybody else is furious. They never seem to learn. Of course they'll kick a neighbour who is down.
Yes and no. Most of the Golan Heights were grabbed during the Six-Day war, which was one of a series of wars - and ones where the Arabs wanted to grab Israeli territory. In fact, in some of the wars they wanted to destroy Israel.
Also Israel has been willing to hand land back in the past: the Sinai being an extremely large example.
So it is more complex than that. But on the other hand, the settlers...
The point here, though, is that the border with Syria (however disputed) has been stable for fifty years. Folk can go on as much as they like about a 'state of war', but this is not a trivial development.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
However, random people in their back bedroom can't train LLMs and even with a few nVidia 9090s in 2040 you won't be able to. We aren't going to get from needing 100k GPUs and billions a year on compute costs to home usage in the foreseeable future. So if they put in place a whole amount of regulation, you won't be able to train SOTA models nor will small startups and they will lag behind the authorised ones which will be a small number of companies.
Now could some people train some small models on something specialist yes, and open source them, but its a bit like you can hack games consoles, people do, can you buy Raspberry Pi and do emulation, sure, but Sony, Microsoft and Nintendo are the only ones making them for the masses. And as we see with emulation, the big companies absolutely monster these people making open source emulation software and tie them up time and time again with legal attacks.
The obvious way this gets shut out to the masses is that huge chunks of the training data is protected, its theft to use it, and only these small number of regulated companies are allowed to use it...the fact they got there by "acquiring" it will be seen as simply them innovating to start with and now they are very sorry and they make sure they make big payments for the right to use it.
In fact, running on home computers removes the cost and liabilities of running the model for you, from the vendors.
Microsoft will have an advantage, I think, in the medium temp, if local models take off. They can afford to buy a top notch model and bundle it with a 365 license - they have a vast captive audience for that. A few quid on the price will barely be noticed. They also have plenty of experience in securing their stuff against copying.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
However, random people in their back bedroom can't train LLMs and even with a few nVidia 9090s in 2040 you won't be able to. We aren't going to get from needing 100k GPUs and billions a year on compute costs to home usage in the foreseeable future. So if they put in place a whole amount of regulation, you won't be able to train SOTA models nor will small startups and they will lag behind the authorised ones which will be a small number of companies.
Now could some people train some small models on something specialist yes, and open source them, but its a bit like you can hack games consoles, people do, can you buy Raspberry Pi and do emulation, sure, but Sony, Microsoft and Nintendo are the only ones making them for the masses. And as we see with emulation, the big companies absolutely monster these people making open source emulation software and tie them up time and time again with legal attacks.
The obvious way this gets shut out to the masses is that huge chunks of the training data is protected, its theft to use it, and only these small number of regulated companies are allowed to use it...the fact they got there by "acquiring" it will be seen as simply them innovating to start with and now they are very sorry and they make sure they make big payments for the right to use it.
In fact, running on home computers removes the cost and liabilities of running the model for you, from the vendors.
Microsoft will have an advantage, I think, in the medium temp, if local models take off. They can afford to buy a top notch model and bundle it with a 365 license - they have a vast captive audience for that. A few quid on the price will barely be noticed. They also have plenty of experience in securing their stuff against copying.
Any idea what the memory requirement is for Robert's model ?
Unsurprising. The Israelis have been nicking other people's land for a century. The weird thing is how they then act surprised when everybody else is furious. They never seem to learn. Of course they'll kick a neighbour who is down.
Yes and no. Most of the Golan Heights were grabbed during the Six-Day war, which was one of a series of wars - and ones where the Arabs wanted to grab Israeli territory. In fact, in some of the wars they wanted to destroy Israel.
Also Israel has been willing to hand land back in the past: the Sinai being an extremely large example.
So it is more complex than that. But on the other hand, the settlers...
The point here, though, is that the border with Syria (however disputed) has been stable for fifty years. Folk can go on as much as they like about a 'state of war', but this is not a trivial development.
I agree.
Incidentally, I read yesterday that Assad and Israel had a security agreement, and Israel are worried that the new regime will not do the same. In which case, they're pretty much ensuring the new regime will not...
Jury "nullification" must be a risk for the prosecution of the CEO killer. Just astonishing levels of support, with the McDonald's where he was apparently reported to the police swamped with negative reviews (it was a "boomer patron" according to the internet).
One of the most interesting things to come out of the US for some time, particularly in the context of Trump's election. People keep on talking about wealth inequality being higher in the US now than it was at the start of the French Revolution...
Very disturbing that so many people appear to support the worse crime that anyone can commit.
I'd argue that it was far from the (worst?) crime anyone can commit. It was cold-blooded murder, but at least it was quick and efficient.
How does that compare to (say) a father raping a child? Or a decades-long abusive relationship between husband and wife? Or the stories of torture and murder we have seen coming out of Sednaya Prison? Or a gunman running amock in a school, killing at random?
The guy did an evil, but it is by far the deepest evil.
(I am not defending what he did; he should go to jail for life.)
Mangione should certainly go to jail. But, Brian Thompson should have gone to jail, too. He was, essentially, practising fraud on his company’s policyholders.
asks people who they prefer out of Scholz and Merz to be next chancellor Scholz 43% Merz 45%
That Merz is only 2% ahead of the widely despised Scholz, who is leading a historically unpopular government, and whose party (on 15% in this poll) is almost 20% behind CDU/CSU in the polls, shows how unconvinced people are by Merz.
Unsurprising. The Israelis have been nicking other people's land for a century. The weird thing is how they then act surprised when everybody else is furious. They never seem to learn. Of course they'll kick a neighbour who is down.
Yes and no. Most of the Golan Heights were grabbed during the Six-Day war, which was one of a series of wars - and ones where the Arabs wanted to grab Israeli territory. In fact, in some of the wars they wanted to destroy Israel.
Also Israel has been willing to hand land back in the past: the Sinai being an extremely large example.
So it is more complex than that. But on the other hand, the settlers...
The point here, though, is that the border with Syria (however disputed) has been stable for fifty years. Folk can go on as much as they like about a 'state of war', but this is not a trivial development.
I agree.
Incidentally, I read yesterday that Assad and Israel had a security agreement, and Israel are worried that the new regime will not do the same. In which case, they're pretty much ensuring the new regime will not...
Yep.
Obviously the new regime has to be judged by its actions. Pre-empting that risks creating an enemy state unnecessarily.
More than that, it's not impossible (however unlikely you think it might be) that constructive engagement might create a friendly neighbour.
Despite all the hype, all the high profile people saying they are leaving etc, after a burst new users in November, it hasn't really gone all hockey stick, up another few million over the past month and still only 24 million user accounts (its creeping up, but that is still tiny in the grand scheme of things) and I have seen stats that show activity daily has dipped in the past couple of weeks.
I saw a chart online showing sign ups were pretty much back to what they were before the flounce/exodus from Twitter.
asks people who they prefer out of Scholz and Merz to be next chancellor Scholz 43% Merz 45%
That Merz is only 2% ahead of the widely despised Scholz, who is leading a historically unpopular government, and whose party (on 15% in this poll) is almost 20% behind CDU/CSU in the polls, shows how unconvinced people are by Merz.
Does this point to the possibility of a serious polling miss when the time comes?
Jury "nullification" must be a risk for the prosecution of the CEO killer. Just astonishing levels of support, with the McDonald's where he was apparently reported to the police swamped with negative reviews (it was a "boomer patron" according to the internet).
One of the most interesting things to come out of the US for some time, particularly in the context of Trump's election. People keep on talking about wealth inequality being higher in the US now than it was at the start of the French Revolution...
Very disturbing that so many people appear to support the worse crime that anyone can commit.
I'd argue that it was far from the (worst?) crime anyone can commit. It was cold-blooded murder, but at least it was quick and efficient.
How does that compare to (say) a father raping a child? Or a decades-long abusive relationship between husband and wife? Or the stories of torture and murder we have seen coming out of Sednaya Prison? Or a gunman running amock in a school, killing at random?
The guy did an evil, but it is by far the deepest evil.
(I am not defending what he did; he should go to jail for life.)
Mangione should certainly go to jail. But, Brian Thompson should have gone to jail, too. He was, essentially, practising fraud on his company’s policyholders.
It wasn't achievable new innovation in insurance, though his use of AI to deny claims might be.
Wasn't the daytime job of Mr Incredible denying insurance claims?
Must have been expected (Holly is a big Nige supporter) but still a blow.
I might be wrong, but I think he stopped giving to the Tories a while a go*, no? He has a lot of money and not afraid to donate it.
* lots of their usual donors went on strike in run up to the GE.
UKIP, Reform etc are all just spin offs to drag the Conservatives to the hard right, the multi-millionaire donors and Tufton street think tanks will happily swap back when they want some of the respectability that the Conservative party retains.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
However, random people in their back bedroom can't train LLMs and even with a few nVidia 9090s in 2040 you won't be able to. We aren't going to get from needing 100k GPUs and billions a year on compute costs to home usage in the foreseeable future. So if they put in place a whole amount of regulation, you won't be able to train SOTA models nor will small startups and they will lag behind the authorised ones which will be a small number of companies.
Now could some people train some small models on something specialist yes, and open source them, but its a bit like you can hack games consoles, people do, can you buy Raspberry Pi and do emulation, sure, but Sony, Microsoft and Nintendo are the only ones making them for the masses. And as we see with emulation, the big companies absolutely monster these people making open source emulation software and tie them up time and time again with legal attacks.
The obvious way this gets shut out to the masses is that huge chunks of the training data is protected, its theft to use it, and only these small number of regulated companies are allowed to use it...the fact they got there by "acquiring" it will be seen as simply them innovating to start with and now they are very sorry and they make sure they make big payments for the right to use it.
In fact, running on home computers removes the cost and liabilities of running the model for you, from the vendors.
Microsoft will have an advantage, I think, in the medium temp, if local models take off. They can afford to buy a top notch model and bundle it with a 365 license - they have a vast captive audience for that. A few quid on the price will barely be noticed. They also have plenty of experience in securing their stuff against copying.
Any idea what the memory requirement is for Robert's model ?
Oh, lots.
The next benchmarks for new machines, will be what can it run locally and how fast.
I'm using the Llama3.2 billion parameter LLM model running locally on my PC, and *man* it's good. I'd go further, it's extraordinarily good. I'm using an AMD mini PC with integrated graphics, and I'm getting performance that is roughly equivalent to GPT3.5 from six or nine months ago.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
Which is why all the people who've invested in AI were so keen to warn about the dangers of the tech: they were worried that they had no magic secret sauce, and that anyone could replicate what they were doing, given enough (stolen) data.
Which is why the large incumbents are warning of the “dangers” and pushing for regulation which creates significant barriers to entry.
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
How won't I be able to?
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
However, random people in their back bedroom can't train LLMs and even with a few nVidia 9090s in 2040 you won't be able to. We aren't going to get from needing 100k GPUs and billions a year on compute costs to home usage in the foreseeable future. So if they put in place a whole amount of regulation, you won't be able to train SOTA models nor will small startups and they will lag behind the authorised ones which will be a small number of companies.
Now could some people train some small models on something specialist yes, and open source them, but its a bit like you can hack games consoles, people do, can you buy Raspberry Pi and do emulation, sure, but Sony, Microsoft and Nintendo are the only ones making them for the masses. And as we see with emulation, the big companies absolutely monster these people making open source emulation software and tie them up time and time again with legal attacks.
The obvious way this gets shut out to the masses is that huge chunks of the training data is protected, its theft to use it, and only these small number of regulated companies are allowed to use it...the fact they got there by "acquiring" it will be seen as simply them innovating to start with and now they are very sorry and they make sure they make big payments for the right to use it.
In fact, running on home computers removes the cost and liabilities of running the model for you, from the vendors.
Microsoft will have an advantage, I think, in the medium temp, if local models take off. They can afford to buy a top notch model and bundle it with a 365 license - they have a vast captive audience for that. A few quid on the price will barely be noticed. They also have plenty of experience in securing their stuff against copying.
Any idea what the memory requirement is for Robert's model ?
Oh, lots.
The next benchmarks for new machines, will be what can it run locally and how fast.
Um to run Llama 3.2 1b only requires 8gb of memory
asks people who they prefer out of Scholz and Merz to be next chancellor Scholz 43% Merz 45%
That Merz is only 2% ahead of the widely despised Scholz, who is leading a historically unpopular government, and whose party (on 15% in this poll) is almost 20% behind CDU/CSU in the polls, shows how unconvinced people are by Merz.
That is probably because Merz would be the most rightwing German chancellor since WW2. No centrist like Merkel is he.
The average SPD leader would beat him, the fact the SPD is running so unpopular a government is what gives Merz the chance to take power
Comments
Off topic. Israel is reported to have launched about 150 'airstrikes' on targets in Syria. We don't know what that 150 relates to; launches, separate targets or sorties but any way you look at it, its a lot. Notably strikes have occurred at the old Russian naval facility at Latakkia. The indictions are that the current Syrian government is open to allowing Russia to retain a presence but Israel might conclude that it needs to lob some more at those former facilities so they arent easy to walk into. Ostensibly these strikes are designed to take out chemical weapon related facilities and heavy missiles but, they are also attacking the likes of air defence capabilities and storage sites. This suggests part of the plan is to support Israels ability to act with impunity in Syria if it needs to.
Meanwhile a number of hundred Russian troops and military contractors are a bit stuck at the moment. Russia has asked the Turks to help out and the likelihood is they will get out ok, but its a worry given some serious local hostility.
When Chump was up for his E Jean Carroll charges, it was all about identifying a single juror who followed only Trumpish news sources, in the hope that there would be one nail-Juror.
https://www.theguardian.com/media/2024/dec/09/rupert-murdochmedia-empire-children
Please don't tell the trees this; they might go on strike.
(Freeman Dyson has some interesting thoughts on this problem,)
https://www.youtube.com/watch?v=hVNjU2NhDDI
Personally, I think we should just relax about the use of gas. It'll get eventually displaced by solar, but in the interim it's better than most of the alternatives.
Would you like to have something GPT4 quality running on your own computer (assuming you have a decently powerful computer)?
Here you go: https://simonwillison.net/2024/Dec/9/llama-33-70b/
Also, a reminder of just fori good the open source LLMs are getting.
I am saying that the permanent seizure of territory is completely unjustified.
It's very pretty clear from Israeli cabinet statement thaf is what is intended.
It potentially cuts off both Iranian and Russian influence in Syria, and the weapons supplies that have been used against them.
By immediately attacking the new regime, they risk what might have been an end to active hostilities with Syria. And potentially undoing that gain.
PM’s words follow Israeli takeover of previously demilitarised zone in Syrian-controlled territory
https://www.theguardian.com/world/2024/dec/09/israel-seizes-syrian-buffer-zone-amid-airstrikes-on-regime-weapons-depots
..
Now, either Bluesky is going forward or it's going backward. But right now, it is very definitely going forward.
I also suspect that it will never end up with a 100% share, like Twitter used to have. The internet is fragmenting, and people are choosing to only see things they agree with. Bluesky will therefore never hockeystick to 100% of the "twitter"-type space, because it only appeals to the (substantial) minority of people who want Twitter and hate Elon.
BUT THIS IS RUNNING LOCALLY ON MY (inexpensive) PC, and it's not even causing the fan to turn on.
That's extraordinary to me.
But there's an awful lot of open source data out there, including much of the open web. If you scrape the whole of public GitHub, you will know almost all you could ever need to know about coding.
Now, there are incumbents. Bluesky is not operating in a vacuum; it is an insurgent against a massive competitor. And that's far more difficult than trying to build something where, at worst, your competitors are the same size as you.
It is much harder to go hockey-stick against a massive incumbent. The fact Bluesky has taken off as it has is either a sign they're brilliant, and/or that Twix is mucking up badly.
Also, they were not small. Less than a year after launching, they spent a billion dollars buying a Yankian firm. They had very big Chinese backers.
If your code is open source and you dumped it in Github, then that's not stolen.
You chose to make it available under an open source license, and I don't really see why people shouldn't be allowed to use it to train their model.
But this is all irrelevant.
There's more than one way to skin a cat. Some services will go hyperbolic. Others will be slower constant growers.
And most - of course - will fail.
There's no immutable law that says that if you're growing at 10% a month rather than 25%, then you're doomed.
Look at mobile gaming, there are plenty of examples of games which went 1, 100, 10,000, and then all the way back to 1 very quickly. While other games, have grown much more slowly, but have shown much greater sticking power.
https://www.skeletonclaw.com/
https://www.bbc.co.uk/news/articles/cz6l1xjnd6xo
e.g.: https://www.vice.com/en/article/ai-is-probably-using-your-images-and-its-not-easy-to-opt-out/
You won’t be allowed to spin up your own open source AI and use it commercially, in the same way you can’t just rent a unit in the mall and set up a bank. But Microsoft/Google/Meta/Apple will all rent you one for $10 a month, totally compliant with all those hundreds of pages of applicable regulations.
The Racing Minister, Winston Peters, played the “it’s all about the welfare of the dogs” card and it’s probably supported by the majority over here. More than 1,000 people will lose their jobs but the tracks are candidates for redevelopment and that will bring in some money to soften the blow.
The last dog meeting will likely be in July 2026. That will leave only the UK, Ireland, Australia and parts of the USA. Apparently it’s legal in Mexico and Vietnam but there are no tracks - as to whether there are any dogs, I’ll leave that to others.
I have always said, even before Musk, twitter as a product is crap and they crucially (unlike Meta) don't know their customers, which always makes selling ads harder and less valuable. Before Musk, there was virtually no feature development.
This reminds me of when governments were going to ban encryption of beyond a certain complexity. And they did. And open source algorithms existed, and were universally used.
The laws were out of date and simply ignored.
Once something has been learned, it cannot be unlearned.
The underpinnings of deep learning and LLMs are advancing far faster than regulation can constrain them.
But I think you make a much weaker case regarding open source code.
Not ignoring people. Just sleeping.
Now could some people train some small models on something specialist yes, and open source them, but its a bit like you can hack games consoles, people do, can you buy Raspberry Pi and do emulation, sure, but Sony, Microsoft and Nintendo are the only ones making them for the masses. And as we see with emulation, the big companies absolutely monster these people making open source emulation software and tie them up time and time again with legal attacks.
The obvious way this gets shut out to the masses is that huge chunks of the training data is protected, its theft to use it, and only these small number of regulated companies are allowed to use it...the fact they got there by "acquiring" it will be seen as simply them innovating to start with and now they are very sorry and they make sure they make big payments for the right to use it.
Talking of banning maths, the recent quantum computing developments are, if replicable, potentially even more world-changing.
Also Israel has been willing to hand land back in the past: the Sinai being an extremely large example.
So it is more complex than that. But on the other hand, the settlers...
Quantum Computing: Tech's Longest-Running Hoax
https://www.youtube.com/watch?v=xcbZJDJlptk
Google have repeated been caught been rather naughty this this stuff in the way they benchmark things.
https://order-order.com/2024/12/09/tory-donor-nick-candy-joining-reform-as-party-treasurer/
Must have been expected (Holly is a big Nige supporter) but still a blow.
How does that compare to (say) a father raping a child? Or a decades-long abusive relationship between husband and wife? Or the stories of torture and murder we have seen coming out of Sednaya Prison? Or a gunman running amock in a school, killing at random?
The guy did an evil, but it is by far the deepest evil.
(I am not defending what he did; he should go to jail for life.)
* lots of their usual donors went on strike in run up to the GE.
One paper isn’t a revolution, and a quantum computer has always been about as far away as cold fusion.
F1: won't be for a little while but when I put up a post-season betting review the graph is going to be rather lopsided. I was behind almost the entire season until the last half dozen races, all but one of which was green.
https://bsky.app/profile/evgen-istrebin.bsky.social/post/3lcwm2w37vc24
At what point does this become hyperinflation?
Post-season test taking place in Abu Dhabi, with each team using one of the regular drivers for tyre testing, and a young driver.
https://www.pistonheads.com/gassing/topic.asp?h=0&f=&t=2101447
https://www.autosport.com/f1/news/f1s-abu-dhabi-post-season-test-who-is-driving-in-2024-and-more/10679999/
Which might be a good thing as all the mood music last season was how terrible Haas were going to be.
Let’s also get that gas pipeline built through Syria, which will strangle them even more.
Interest rates are 20% as well, and the rouble lost 25% against the USD between August and last week, since when it’s bounced a little.
(Though again AIUI, they did the same thing last year, and the ruble's value recovered when trading resumed.)
Folk can go on as much as they like about a 'state of war', but this is not a trivial development.
Microsoft will have an advantage, I think, in the medium temp, if local models take off. They can afford to buy a top notch model and bundle it with a 365 license - they have a vast captive audience for that. A few quid on the price will barely be noticed. They also have plenty of experience in securing their stuff against copying.
https://www.theguardian.com/music/2024/dec/09/the-gates-of-heaven-rod-stewart-to-upgrade-his-world-class-model-railway
Incidentally, I read yesterday that Assad and Israel had a security agreement, and Israel are worried that the new regime will not do the same. In which case, they're pretty much ensuring the new regime will not...
https://www.theguardian.com/world/2024/dec/10/who-is-friedrich-merz-germany-next-chancellor-scholz-merkel
which conspicuously fails to mention how very unpopular Merz is.
For example, this poll from a couple of days ago
https://www.zeit.de/politik/deutschland/2024-12/olaf-scholz-friedrich-merz-zdf-politbarometer-fast-gleichauf
asks people who they prefer out of Scholz and Merz to be next chancellor
Scholz 43%
Merz 45%
That Merz is only 2% ahead of the widely despised Scholz, who is leading a historically unpopular government, and whose party (on 15% in this poll) is almost 20% behind CDU/CSU in the polls, shows how unconvinced people are by Merz.
Obviously the new regime has to be judged by its actions. Pre-empting that risks creating an enemy state unnecessarily.
More than that, it's not impossible (however unlikely you think it might be) that constructive engagement might create a friendly neighbour.
And Twitter sign ups were pretty stable.
Sadly I cannot find it.
NEW THREAD
Wasn't the daytime job of Mr Incredible denying insurance claims?
The next benchmarks for new machines, will be what can it run locally and how fast.
The average SPD leader would beat him, the fact the SPD is running so unpopular a government is what gives Merz the chance to take power