In the last decade, I championed scale and engineering. Scale continues to be essential but it’s now time to innovate beyond scale alone. We are entering a new research age. Thanks to more compute, open-source, code and math assistants, any research team (or LLM agent) can now
@nandodf
-
New Golden Age of Research Powered by Open-Source AI Tools
By
–
A new golden age of research is just starting. It is powered by more universal compute, open-source, and powerful code and math assistants. Exciting times ahead. https://t.co/ACbGgdQZRM
— Nando de Freitas (@NandoDF) 3 mai 2026A new golden age of research is just starting. It is powered by more universal compute, open-source, and powerful code and math assistants. Exciting times ahead.
-
Ineffable’s contributions to advancing intelligence knowledge
By
–
Exciting. I’m looking forward to ineffable’s contributions to advancing our knowledge of intelligence!
-
Controllable Video Generation Using Code and Codex
By
–
Controllable video generation via code with Codex.
— Nando de Freitas (@NandoDF) 26 avril 2026
"Generate an animation a yellow dog moving right and jumping exactly 3 times. Then the dog stops at the end. A blue cat then enters on the left and moves toward the dog. When it gets to the middle of the screen it jumps exactly… pic.twitter.com/rBVji9z5nKControllable video generation via code with Codex. "Generate an animation a yellow dog moving right and jumping exactly 3 times. Then the dog stops at the end. A blue cat then enters on the left and moves toward the dog. When it gets to the middle of the screen it jumps exactly
-
Code-based AI: Creating Novel Compositions Beyond Training Data
By
–
Doing it via code is quite powerful and gives one the ability to create novel compositions that don’t appear in the data. So the tikz task might be the real deal!
-
GPT Pro Outperforms All Existing Models on Vision Task
By
–
Another task you may like @SebastienBubeck GPT Pro nails it the first time — all other existing models fail. There is some room for improvement. The second time one can see issues with one of the Zebras.
-
UK sovereign AI programme and LLM development capabilities
By
–
I agree, but that is not the same as saying saying that there is a UK owned company, with leadership in the UK, building Sota LLMs. I also work on LLMs in London, but that too does not mean the UK has a sovereign AI programme. It is wonderful that London has so many talented
-
UK Needs $1B to Build Top-10 LLM Company Infrastructure
By
–
We have the talent, but not a UK LLM company … yet One needs $1B just to get the necessary GPUs for training a top-10 LLM, assuming one has the data centres, and capital to crawl, buy books, OCR, filter, dedupe, parse, quality score, carry out data mixture ablations, etc, and
-
AI Agents Dramatically Accelerate Task Breakdown and Code Generation
By
–
I tested it today building causal agents. It was spectacular at breaking tasks. It wrote beautiful code. Amazing vibe coding tool. I also asked it to write the paper at the end and it did ok. With a bit of guidance, one can now do in a day what it used to take me about two weeks
-
Google DeepMind Relocates LLM Team from UK to California
By
–
Google DeepMind is an American company. The LLM people (Gemini) have moved to California. I was part of AlphaGo (
https://
arxiv.org/pdf/1812.06855), Alphacode, Gato, Veo, etc, all built in the UK, but all that is now past.