north of noise

naturalist and nerd

[Ross Taylor @rosstaylor90]

A few words on the Sovereign AI debate, having built several LLMs in Meta while in the UK and now working as a UK based startup:

1. Lots of people are trying to do the right thing to make the UK a better place to start AI companies. Time lags until the benefit show, but you should judge on the intent now. I support the direction of travel!

2. DeepMind has been enormously beneficial for the UK, but it has muddied the waters for a sovereign LLM company to emerge as (until recently) the Government continued to celebrate it as a British achievement / push it as a national champion.

3. Similarly, people are now celebrating recent US investment in King’s Cross, while also wanting more UK sovereignty. Clearly some income effects here, but I would worry about the substitution effects too. AI is not like other types of foreign investment.

4. The relevant talent nexuses in UK that could develop a competitive foundation model are from GDM and old Meta AI GenAI. Also some folks from smaller groups, ex Conjecture, Stability. The talent is still there, although a lot was snapped up by US FM companies in the past year. I personally think it’s not too difficult to develop new talent either from UK universities, but you probably need an ex GDM or Meta core (Gemini or Llama). Or if not: show evidence first (technical reports) before claiming you can do it.

5. Building an LLM is very different from doing regular AI research - skillset is different. Former is closer to engineering; long hours, often unsexy work. Important to distinguish between these two types of talent in the UK ecosystem; arguably too much focus on the latter / ideas guys.

6. On research - DeepSeek R1 post-train cost $300k . Yes, they also needed an ablation budget and to train a base model, invest in infra and talent - and yes the cost of an R1 moment is increasing year on year - but the idea that you need $1bn plus immediately to show results is complete FUD. You need billions to scale, not to validate new directions.

7. In my experience, every failed LLM effort (from model results perspective) I witnessed in the past came from a combination of poor leadership, politics, unclear vision, and premature scaling. Good efforts usually started from small teams who had worked with each other for a long time, had shared thesis, and scaled progressively in bite-sized pieces. Some recent lessons here for neolabs as well.

8. Things take time. Eg we’ve spent ~12 months mostly on internal infra just to get into the position to be able to make big swings. It’s important to nurture new companies through the initial phase. Expectation management is also crucial. I think expecting new UK companies to have single big bang releases is very dangerous; sort of like overwatering a plant. The correct release pattern is “decent”. “decent”, “decent”, “quite good actually”, “holy shit”.

9. Please don’t allow politicians or journalists to kill recent or upcoming AI investment efforts. We will need way more - at the price of potential inefficiency in places - as AI is existential for the country. Ambitious projects are usually incredibly fragile in the early stages; look after them!

10. Mythos is a good triggering moment, but what’s coming will make it look like a toy, so it’s worth building for what’s coming in 5 years time - not a current generation model.

Very proud to be building in the UK - more to share on that soon - alongside many other great early stage AI companies! 🇬🇧

[north of noise @euxoa]

@Afinetheorem oh i didn’t actually know it’s totally disabled

[north of noise @euxoa]

@ObhishekSaha @slatestarcodex he’s a psychiatrist

[north of noise @euxoa]

@AndrewCurran_ gradually though, over a longer timeline (i believe), while the thin loop will become faster

[north of noise @euxoa]

@AndrewCurran_ Yes, that’s basically autonomous technology (the process, not a single capability) and it will come.

[north of noise @euxoa]

Yes, i see the loop and it’s bad news for competition, apparently now including all “middle powers” and China who really need to get their shit together.

But anyway, I don’t see the loop self-sustaining in a way that would allow an incompetent party to just operate it, which would be the extreme case with a nationalization scenario.

We may have a bit different idea about the loop in general. I see it more like @tszzl: the loop has been around at least since renaissance, in a way. While it will have a new tight core, it is still strongly dependent on its broader organizational and economic context.

I really do hope this looming mess can be sorted out peacefully. Right now it looks a bit dark. Say China for example, if an RSI in the strong sense happens, they will not just stay on the sidelines and watch Taiwan delivering the ingredients.

[north of noise @euxoa]

There’s a bit of self-sabotage built into the latest US government actions on AI, as it erodes trust of foreign users, aliens potential recruits, and disables a major part of the researchers in a frontier lab.

Overall a net loss, but if we need to go to AI nationalism anyway, an incompetent US government and an early warning is not an entirely bad thing, from the non-US point of view.

[tylercowen @tylercowen]

A few thoughts on the recent Mythos brouhaha: https://t.co/v25AJdeViy

[antirez @antirez]

What Europe should do right now:

1. Call all the European researchers working on AI and return them back with same salary (or they can stay but switch career).

2. Fill EU places having GPUs with money, and put those people there.

3. AI partnerships with China + India.

[Sam Bowman @s8mb]

This is, perversely, good news for Britain, Australia, Japan, Europe, and other countries being cut off that would once have seen themselves as close allies of the United States.

It shows us what the future may hold if AI is the strategically and economically decisive technology of the 21st century and is controlled by the US and China. It is good news because *it may be happening early enough to give us time to act.*

I think this will be rescinded pretty soon, but it’s a sign of things to come. In a future where frontier models cannot be used outside the US, our industries and economies will fall behind and American businesses may not be able to operate overseas. We won’t be able to defend ourselves militarily with defence systems built on obsolete software. Europe 2031 is a good scenario of what a future like this could mean: https://t.co/AMc5LrFJeS

Some of the things we need to do are ‘no regrets’ measures we should do anyway. But some are genuinely costly and risky.

We need cheap electricity – powered by gas, coal (this is costly, coal is very bad), deregulated nuclear fission – whatever can provide *cheap, reliable, 24/7* power. This almost certainly excludes wind power, which is enormously expensive and unreliable. We need projects to be able to connect to the grid in days rather than years by paying for fast-track connections.

We need to make it incredibly easy to build data centres, with the property taxes retained locally and hypothecated for local tax cuts so there is some direct benefit for locals. This doesn’t need to be nationwide.

We need to create new regulatory regimes for innovative businesses that give them the right to hire and fire staff with ease. The difficulty and cost of firing staff is one of the main reasons Europe has fallen behind so badly. We need to create a parallel employment regime that companies and workers can opt in to: https://t.co/YaNOXK1Po2

Even though I think it will probably fail, I think we should probably try to create a good, non-American frontier AI lab. I am quite pessimistic about this – even extremely well-resourced, innovative software companies are struggling to do this. But the stakes are so high that not trying seems foolish.

One thing that might work in our favour is the number of brilliant AI engineers who are not US citizens, who under the current export controls do not have access to Mythos/Fable even if they live and work in the US. What happens to Demis Hassabis, Ilya Sutskever, Andrej Karpathy, and the many other Europeans, Canadians, etc who are working on AI models in Britain and America who are affected by this?

I do not think we should force our own companies to use model, because this would exacerbate their economic weakness – this lab should have to compete on an even playing field. I am deeply sceptical that this can work, but we cannot rule it out. If we do it, it has to be able to pay US salaries, operate without political constraints. https://t.co/Um05rUF4Vq

It is cope to tell yourself that Trump is an aberration or that these export controls are a one-off. To repeat, I think these specific controls will be lifted quickly and it will be easy to move on and forget it happened. But this is a look into a potential future. Every one of us that is not a US citizen is at risk. The standard political divides do not apply here; the question is whether you grasp the enormity of AI as a technology. We have to act!

[north of noise @euxoa]

(Assuming your #2 here.) This is supposed to be temporary, we'll see. The whole Mythos/Fable availability has been a mess. If the same policy instincts are followed in future, all models of the Mythos class or better will be restricted by default, and released only selectively, either by functionality or by user, kind of semi-nationalized.

I wonder if EU wakes up. And what will China do in the longer run.

[Samuel Hammond 🦉 @hamandcheese]

Two possibilities, both extremely bad:
1. The admin is deliberating fucking with Anthropic for Schmittian reasons
2. This is now precedent for any model with Mythos-level capabilities and beyond

When OpenAI releases a comparable model will they face the same controls?

[roon @tszzl]

rsi is a process that’s been happening at least since the renaissance

[Sherpa @LLMSherpa]

Folks won't shut up about Fable 5, but Gemini is out here SAVING LIVES.

[north of noise @euxoa]

@bakkermichiel I tried to share this at work, but Palo Alto has security filters that block the site. ;)

[Michiel Bakker @bakkermichiel]

"What will happen to Europe if it keeps ignoring AI?"

Three American labs each (!!) operate more AI compute than all of Europe combined. Today we're launching Europe 2031: a story of what might happen if that doesn't change.

[north of noise @euxoa]

@claudeai Fix your safety filters. For me, Fable is just unusable for any sort of casual chat. I can't even get a feeling of the model because I'm constantly switched back to Opus. My chats are not even remotely related to any of the areas of security concerns.

[Avraham Eisenberg @avi_eisen]

Is this AGI