Oh god, the @aidungeon flashbacks! /cc @vrandezo

Also: UK-Japan Joint Declaration on Economic Security Cooperation gov.uk/government/news/uk-ja…

Substantial: 🇬🇧 and 🇯🇵 sign a Frontier Technology Partnership. Builds on/combines “the UK’s world-class software and research leadership with Japan’s unparalleled hardware and manufacturing power to support collaboration across our distinct yet highly compatible industrial bases.” AI • “UK and Japan will be AI makers and not just AI takers, fostering resilient, safe, secure, and trustworthy AI ecosystems and enhancing our national AI capabilities.” • Joint science and semi research and formal engagement. • UK and Japan AISIs to work more deeply together. Quantum • Co-develop globally competitive, commercially scalable and deployable quantum technologies, including computing, sensing and communications, building on the Quantum Memorandum of Cooperation (2025). Biorisk • Deepen efforts to counter deter novel and emerging biological threats as part of our broader approach to dual-use S&T, strengthening shared commitment to non-proliferation, and strategic approaches to enhancing biological security as part of wider cooperation on science and technology. Emerging tech • We will work together to protect critical and emerging technologies through deepening cooperation on research security, recognising the need to manage risks associated with those technologies, while supporting open, secure, and trusted international research collaboration. • We commit to working bilaterally and with like-minded partners to share information on policy measures to reduce loss of critical technology. This work will be complementary to the Joint Declaration on Economic Security Cooperation.

This is a *way* bigger deal than it seems... Frontier AI companies will *never* own the frontier again I kid you not... I've been waiting for someone to show this result for like 4 years... this is a huge deal. The short reason: combinations of models will *always* outperform individual models The long reason: this is the gateway to a million times more data... and huge leaps in compute efficiency. The AI scaling laws always win. More in article below 👇

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! 🇬🇧

Huh, we didn’t have this in the scenario until 2029 europe2031.ai/

UK AI Minster @KanishkaNarayan says when testing Anthropic's Fable 5 it was the first model ever to get through their entire cyber-range with no human support: "The cyber range is supposed to be what any company's usual cybersecurity network, tries to stop and that model was able to get through that end to end with no human support." "The fact that we're able to do this way before release and then through the National Cyber Security Centre mitigate the risks for critical national infrastructure is unique." "No other country in the world outside of the United States... has that capability and that access."

I want to explain why the Defence world is devastated at John Healey's resignation. In every other Dept of State, the worst thing imaginable is manageable - pandemic, flood, riot, even a bond market selloff. Defence people prep, 24/7 for the unimaginable. Today someone stood up for Defence - not "against" HMG but against the public complacency that's limiting HMG's options. No other SecDef has done this... so let's hope this turns the tide.

Matt, I can’t even say “hello” to Fable 5 except in incognito mode (memories off), because it knows I am a biomedical researcher! It would be nice not to ban biomedical scientists before talking access. Isn’t your comment ironic? Let’s first see if you can fix punishing us!

One of the Government’s best ministers doing what a good minister should

We are starting a new, nonprofit alignment organization, ⊢ Sequent Research, bringing together researchers previously on UK AISI’s Alignment Team, Timaeus, and elsewhere to research how to align superintelligence. We are hiring! 🧵

My daughter sends. She knows all about this. Also please come over so I can feed you

This is so funny. If you’re worried about the climate impacts of AI you can buy a hand crank powered AI computer crankgpt.com/

Actually it's fine guys! I figured out a way, see below. Claude Fable 5 is a great model afterall, and I also finally appreciate the difference between CLAUDE.md and AGENTS.md. It's all good.

It is precisely because all movie-making is an exercise in cheating and fakery that there is such a high value placed on “real” filmmaking. The high value parts are the parts the average Joe can’t afford to replicate.

“You’re right to flag it. That clause was an unjustified assumption, and it was doing a lot of load-bearing work in a conclusion I stated as if it were forced.” 🤐

A story about Jon Snow.. We got my Dad's terrible diagnosis thirty years ago while I was working with Jon. We didn't know each other well then but when Jon caught me crying at work, he asked me what was wrong and he listened and was kind. Then...

[1/7] Recent breakthroughs in LLMs’ mathematical ability are genuinely surprising. I recently solved a problem I had been unable to solve for seven years: the optimal acceleration rate for first-order methods under high-order smoothness assumptions in nonconvex optimization.