I always believed in free education. I made a decision to never monetise my popular UBC and Oxford lectures. I’ve lived my life by working to bring knowledge to everyone on Earth: the pioneering lectures, the startups past and new, and my AI research. This desire came as the result of witnessing and experiencing a lot of suffering in my childhood. This course is part of my legacy and I’m very proud of it, and thankful to Oxford, the TAs and my students then.

My friend works at Anthropic. He said they pay $650,000 a year for people who truly understand deep learning. This exact course by Nando de Freitas is free forever. Most AI courses charge $2,000+ for this foundation. This one is completely free. You get the real technical foundation ➜ taught at Oxford by a professor who later joined DeepMind. Save this ⭣

Super happy to release SmolDataEnvs: 5,000 verifiable RL environment tasks for hill-climbing small models in code and data science by @adithya_s_k 100% open source: environments, evals, training! huggingface.co/datasets/Fine…

"JEV-as-a-Judge: Accept When Confident, Escalate When Unsure" This paper shows you can just use JEV for every evaluation instead of expensive LLM. JEV basically acts as a cheap first-pass judge, returning both a verdict and how confident it is. When confidence is high, keep the answer. When it’s low, escalate to a stronger LLM. This simple routing keeps ~99% of GPT-6’s accuracy while reducing evaluation cost by a lot. alphaxiv.org/abs/2609.26550

NVIDIA and Stanford just challenged Jev. (their new System 1 architecture runs up to 9x faster.) It is called a Contrastive Language Model, or CLM. Like Jev, CLM is not designed to generate text. It handles the small, repeated decisions inside AI systems, such as choosing a tool, ranking a patch, routing a request, or selecting the next action. But CLM reaches those decisions differently. Instead of generating an answer token by token, it treats decision-making as a retrieval problem. Here is how it works. 1) Encode the state CLM takes the current situation, such as an agent’s context or the state of a game, and converts it into a vector. It uses a frozen Qwen3-8B model with a small trainable state projection head. 2) Encode every possible action A separate action head converts each candidate into the same vector space. In the Mario example, the candidates are left, jump, and right run. CLM does not invent a fourth option. It only evaluates the actions supplied by the application. 3) Learn which states and actions belong together During training, the correct state-action pair is pulled closer while incorrect pairs are pushed apart. A batch of B examples produces a B × B similarity matrix. The matching pairs sit on the diagonal. Every other pairing becomes a negative example. This contrastive training uses InfoNCE, the same general mechanism behind systems such as CLIP and dense retrieval. 4) Turn similarity into a decision At inference, CLM measures the cosine similarity between the state and every candidate action. A softmax converts those scores into a probability distribution. The application can choose the winner, apply a confidence threshold, or escalate an uncertain result. The real speed advantage comes from separating states and actions. Actions can be embedded once and cached. If an agent repeatedly chooses between the same tools, CLM only needs to encode the changing state and compare it with stored action vectors. That replaces repeated generation with one embedding pass and a set of cheap dot products. The researchers report that CLM-8B matches Jev across computer-use, gaming, and tool-calling evaluations while reaching up to 9x lower latency. The improvement is largest when actions repeat or the candidate set grows. CLM still has limits. It cannot generate new actions, its probabilities are relative to the supplied candidates, and its strongest verifier results require task-specific fine-tuning. But its central idea is powerful. The entire research is open-source, including the code. Read more here: contrastive-lm.notion.site/ When software already knows the possible answers, an AI model should score them instead of generating more words. I also wrote a full breakdown on how system one models like Jev work. The article is quoted below.

Thanks to Ronan Chambers and Luke Knight for an engaging discussion at the @etnshow. It was a real pleasure to talk about @cusp_ai and the societal challenges we can solve through AI powered materials design: nitter.net/i/broadcasts/1nxeLMXpa… at 01:07:11.

This was recorded a few months ago when I was still at Google and it was a really fun conversation!

Highly recommend these opportunities!

Deploying DiffusionGemma-Jev (djev) just got a lot easier. You can now spin up a Jev API-compatible endpoint on Google Cloud Run using a single command. Performance is solid: ~35-60 ms for single step latency and batch@32 is ~100-123 requests/sec. It's a straightforward way to experiment without needing your own GPU. Runs at roughly $3/hr and drops to $0 when idle. Get the code and instructions here: github.com/taeold/djev-run

Can Jev specify graphical models for us?

Inspired by Jev, we built Open-Jev: open-source decision models. 2B/9B: LoRA adapters + decision heads. Code: github.com/Zefan-Cai/Open-Je… Data: huggingface.co/datasets/Zefa… 2B: huggingface.co/ZefanCai/Open… 9B: huggingface.co/ZefanCai/Open… Watch the demo ↓

Jev-as-a-judge🤝 RL Environment Verifiers 🤝 way cheaper + faster RL a lot of things in Agent World look like classification problems…including grading RL - many RL Tasks need a judge to verify pieces of outputs or trajectories - at scale this can be very expensive and slow with tons of rollouts to score - Jev is incredibly fast and cheap meaning, can potentially remove the verification bottleneck for many tasks we previously worked with @harvey on their LAB benchmark which used LLM-as-Judge for tens of criteria per task we found that Harness/Prompt Engineering + Open Models meant we could judge tasks orders of magnitude cheaper like any verifier, it needs to be calibrated to make you’re giving back a good signal but Jev has the opportunity to massively bring down verification costs (and tuning friction) for many tasks which means more teams will do RL, which is great!

UPDATE: Kev-0.6B, 4B, and 8B are now available. Kev is a family of small open source Jev-like decision models you can train and run yourself. This new family is based on Qwen3 using the same LoRA + small pointer head technique as before, but scaled up. Out of domain, on data Kev never trained on: Kev-8B 79.6%, Jev 85.7%. • Drop-in TypeSafe System One API; their SDK works with one `base_url` change • Kev-4B serves on a 32 GB Mac in bf16: ~300 ms for five questions, ~40 ms on an H100 • Repeated documents hit a KV cache: 2-2.5x faster • Apache 2.0 License. Kev-4B trains in 40 minutes on one H100. Kev-8B in 83 minutes. Code, weights, evals: github.com/jaredpalmer/kev

As AI starts making significant dents in science, I think the role of scientists will gradually change. Scientists may spend less time solving problems themselves, and more time deciding which problems are actually worth solving. 🧵 1/X

Wafer just dropped an AI performance engineering repo that covers: > GPU fundamentals & CUDA > kernel optimization > FlashAttention > KV caching, quantization > NVIDIA, AMD & TPU architectures and all resources link to solid docs, papers, and repos.

"SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness" This paper lets coding agents recursively optimize their own harness instead of hand-designing it. Across diverse coding environments, it discovers four mechanisms that compress context, observations, reading, and actions. The result is comparable performance with 44.7-49% less token traffic and roughly one-third lower API cost. alphaxiv.org/abs/2609.20519

This is a fantastic presentation by @CompleteSkeptic — highly recommend it. He is spot on.

We should be developing AI as a tool (1) to empower engineers and scientists to solve the big engineering challenges we face such as energy and the health of our planet (2) to empower more people to code in natural language and ensure the formal code behind the scenes is verifiable, robust, safe and secure, thus creating more jobs (3) to empower people throughout the world with access to general knowledge (4) to empower people with disabilities (5) to empower people suffering with health problems (6) to empower musicians and other artists with tools that enable them to have control and ownership of their data while allowing for them to express themselves in new creative ways (7) to empower historians and other social scientists with their research (8) to empower farmers, and others responsible for our precious food (9) to empower governments and companies with better decision making and planning to deliver better services and welfare … However, agents capable of dangerous tool use, eg autonomous weapons or hacking websites require legal scrutiny. The creators of such agents must be answerable to sovereign laws. Weaponisation of AI is a serious risk for humanity. We need diplomacy. None of the AI goals should ever require that we compromise on safety and human dignity.

🚨 OBAMA ON AI: "If we are thinking about AI just in terms of how do we cure cancer or get better energy, you can do that without having agentic AI and having it just roaming free in the internet. The reason you are doing that is because you have to market a product that people will pay money for. That’s a misalignment between what our society needs and the commercial imperatives that these companies are facing, not because necessarily they’re trying to do bad things, but because they’ve got to justify these valuations. So, that’s one more reason why it is really important for us to have a competent government and a serious bipartisan conversation around this issue, and we have to do it fast. And I would encourage voters to pay attention to this. If somebody does not have a serious plan for how to deal with this, then they’re not meeting the moment, and you should probably look for somebody else."

EXCLUSIVE: Nvidia CEO Jensen Huang tells CBS News' @jolingkent that he does not believe the U.S. should slow down in the AI race against China, but added, "we would never, ever and never should ship products before they're ready." Watch the interview this weekend on @CBSSunday.