Last week we launched Bonsai 2 9B one cuda.fast and mlx.fast. In one week, we improved speed by over 500%. I just wanted to highlight some awesome moments: • ArthurOstapenko set five CUDA records in nine hours on Oct 16, then landed the biggest single jump of the contest the next day: 11 bit-exact kernel changes that fused the drafter, attention, GDN and verify boundary into a bulk-copy ring. 3.67x to 4.16x in 1 submission, with decode reaching 677 tok/s against a 93 tok/s baseline. On MLX, his 4-bit storage for the DFlash drafter's linears produced the largest jump on that board, 3.42x to 3.84x, with the target weights untouched. • kaankolcu made the MLX prefill run on Apple's int8 tensor units by feeding the 2-bit ternary weights int8 activation codes, then later replaced the 15-token greedy draft chain with a 15-node draft tree verified in the same 16-row forward. The final record decodes at 1,229 tok/s on one prompt, 8.5x the stock engine. • spariev took 2 rejected submissions, GordoAR's prompt tile and ArthurOstapenko's attention kernels, showed they sped up different phases, and landed them together with full credit to their authors. Congratulations to everyone who took part and thanks you to @PrismML for giving us Day 0 access a great model!

1/ We discovered something totally unexpected about the neural structure of consciousness that overturns decades of thinking on its biological basis and brings us significantly closer to the recipe for generating conscious awareness. New preprint: biorxiv.org/content/10.64898… (led by @HesseJanis, with @FrankLanfranchi, @_conor_mcgrory, @Yuelin_Shi, and @EngelTatiana)

interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/

My dear friends, I’d like to show you my latest toy project Introducing Handwired: turn your regular JavaScript code into a neural net! - no training - exact outputs vs the JS program - fully differentiable All runnable in a browser, without GPU

DINOv2 has never seen a caption, and Qwen3 has never seen an image. We still aligned their embedding spaces without a single image-caption pair. It even works when the images and the captions come from different datasets. Project page: dominik-schnaus.github.io/un… ⬇️

I don't recommend anyone scramble to move their funds to new wallets today. But we should take the risks to cryptography from AI-accelerated math seriously, and minimize our exposure to not just quantum-vulnerable cryptography, but also potentially AI-vulnerable cryptography. The core new area of risk from this viewpoint is, unfortunately, ML-DSA / FHE / lattices. (and it's also another reason, along with quantum, why ECDSA might fall even faster than expected, hence the "fresh address" recommendation) So far most people have been in the mode of thinking "elliptic curves broken, hashes safe, lattices safe". But there is a good chance that the concrete security of lattices will take serious hits from the next two years of AI math. The basic threat model is: factoring is something that naively takes 2^(n/2) time, but over decades smart people have found and optimized number field sieves, and degraded that to 2^O(n^(1/3)), which is why RSA keys and signatures need to be ~400 bytes (and not 64 bytes). What if there are skeletons in the closet like that, both for elliptic curves and lattices, that we are simply not smart enough to discover - but bots soon will be? This is a major part of the reason why for the past year ethereum's lean roadmap has been going in the "hash-only" direction: no lattices, no ML-DSA, no Falcon, no lattice-based commitments inside ZK proofs, etc. Signatures in lean ethereum are all hash-based, either WOTS or SPHINCS-. For signatures and proofs, we already know how to go hash-only. The bigger challenge is for *public-key encryption* - and this goes far beyond blockchains. Secure communication, anonymizing protocols, lots of things need public-key encryption. And unfortunately there are long-standing mathematical theorems showing why public-key encryption cannot be done with hashes alone. You have to have some kind of trapdoor object that has at least one form of usable "structure" - either group theory (incl. isogenies) or lattices or code-based or potentially in the future even more newfangled and spooky things (local mixing?). But for anything that has structure, you should assume that AI will make at least some progress in breaking that structure. Here, one reasonable inference is that if you want to make something plausibly long-term secure, multiply the key sizes by 10. To me that's a very plausible world and something not at all extreme to predict. If AI will bring us 50 years of math in 2 years, then that 50 years of math may very plausibly include a "naive factoring -> GNFS" level of improvement to our ability to break lattices. In that world, lattices will still exist, but they will have to be significantly bigger to guarantee the same level of safety. And at those new larger sizes, hash-based constructions will beat lattice-based constructions on concrete efficiency in every use case where hash-based constructions are possible at all. Theoretically, of course it's possible that hashes are broken too (eg. P = NP would imply that). But I think P = NP is very unlikely. And intuitively, it's much more likely that a mathematical object has exactly no exploitable structure (like hashes are intended to), than that a mathematical object has exactly ~3 forms of exploitable structure (for elliptic curves: associativity, Schoof, pairings) and not some secret fourth form of structure we have not yet discovered that greatly degrades its security (for elliptic curves, ECDLP and pairing security). Similar for LWE, SVP, RLWE and the zoo of lattice problems. For this reason, we do not yet see any reason to worry and start padding the byte size of hashes (if we start to worry more, we would pad the round count first before doing anything to the byte size). Concrete TLDR, my own personal views: * Hash-based > lattice-based, in those situations where hash-based is possible at all * For anything lattice-based, be much more paranoid on param sizes. Remember that blockchains are only a small portion of the cryptography story; this point goes far beyond blockchains and applies to eg. access to websites, secure messaging, Tor / VPNs ... * For privacy protocols, strongly favor NOT putting encrypted notes onchain. Instead, send them offchain through some third-party mechanism. * If it's not difficult for you, keeping your funds in addresses which have not yet been used to make a transaction is a good idea. If it's easy for you, do it. **But be careful about migrations; I personally have lost more money in botched migrations than I have lost in all hacks combined**. * For multisig wallets, doing confirmations offchain is better than onchain, because this way the signatures of signer wallets do not get exposed to the public, so if ECDSA falls to AI much faster than expected, at least the multisig "gracefully degrades" to a 1-of-1 where the 1 is whoever was gathering the signatures - a much better place to be than "anyone can take the money" firefly.social/post/x/210783…

Is it possible to provably recover individual latent variables of the true world, even without reconstruction (e.g., JEPA)? Yes, with DSReg, a simple regularization that can be applied post hoc to your pretrained model! dsreg.github.io/

LLMs don’t need retraining to become more capable. Our new architecture, Spotlight, gives AI models growing memory, allowing them to gain knowledge and capabilities without changing their weights. Token generation takes constant work, no matter how much information is stored.

Introducing LoopCD our latest research at 🍎 Apple MLR. Heard the rumors that frontier models like GPT-6 Astra and Gemini 4 use Looped Transformers? We present a new inference-time method that makes pre-trained looped models better for (almost) free! AIME 2024 (Ouro-2.6B): 61.9 → 73.3 HumanEval (Huginn): 22.6 → 31.7 No training. No extra model. Read our paper for more details: arxiv.org/abs/2610.02185

Can complex multi-step reasoning emerge purely from cells that only talk to immediate neighbors? Happy to share our paper “Reasoning with Neural Cellular Automata (NCA)”, from our team at Google, Paradigms of Intelligence 🧵👇

Welcome to Kindle Spark OS! TL;DR: Add 4GB of RAM to any DGX Spark & faster! @CK2084, @coffeedev and I figured out how to get 5-10% extra decode and 4GB RAM out of a Spark. We had to replace NVIDIA's OS to do it, but we made it safe and easy. github.com/kindlingai/kindli…

Can an LM, starting from random init (!!), learn to generate all of its pretraining data? Introducing Self-Play Pretraining with Zero Data. Two models start from random initialization: a generator proposes programs for a universal Turing machine and a learner trains on their outputs. We never train on any real data, but see predictable scaling on natural datasets: zero-shot val loss on images, text, audio, and melodies decreases predictably with self-play compute. And the learner develops in-context learning capabilities. A fun proof-of-concept, co-led with @AdityaCowsik and @KfirDolev and co-authors @gbruno_dl, @ANourya @noahdgoodman, and @YoavLevine.

Excited to introduce JEPA-Anything: one predictive principle that works across molecules, cells, fluids, patients, robots, or even weather. A monolithic JEPA target lets dominant signals crowd out weaker modes. Orthogonal Predictive Factorization (OPF) splits capacity across complementary factors you can inspect and test. Code, checkpoints, and the report are fully open. Paper: arxiv.org/abs/2609.20800 Code: github.com/Gen-Verse/JEPA-An…

Bend 2 is here! It is a new programming language that blocks AI mistakes via *proof checking* - the same technique big AI labs used to solve open math problems, like Navier-Stokes. It is also very fast, and runs on GPUs. Watch the video. Link in the comments.

We release Needle 3: A Sliceable 8-29MB automation foundation model that can match DeepSeek V4 Flash. One set of weights, every depth from 2 to 20 layers a model of its own, 25-121M parameters at CQ2-bit, built on our Simple Attention Networks and running locally at up to 4k tokens/sec decode speed on a Raspberry Pi 5. Needle does not chat. Every turn is a function call: give it the tools your app exposes and it picks the right ones and fills every argument from what the user said, or hand it a schema and it returns a typed record. Ask for something no tool covers and you get an empty list, not a guess. That trade is lets 121M parameters trained on 360B tokens of structured data beat models 10x their size on mobile tool calls and match 2-3x bigger models on structured JSON extraction. It runs on mobiles, wearables, smart home devices, small robots and microcontrollers, with prebuilt engines for macOS, Linux, Windows, Android, iOS, watchOS, tvOS, the browser and WASI hosts. Try it in your browser: cactuscompute.com/needle

An extraordinary @Nature paper has just landed from the lab of @Sergiu_P_Pasca at @Stanford - led by Konstantin Kaganovsky. They have found a way to create "xenocortical mice" which have ~92% of their cortex occupied by a cortical organoid derived from human stem cells. This human-derived cortical graft integrates deeply with the host nervous system, supporting organised neural activity and behaviour. This study is a milestone in synthetic biology and chimera research. It has the potential to generate multiple medical breakthroughs by providing an enhanced biological model of human brain tissue that allows behavioural as well as neural and genetic assays. The Stanford team have been exceptionally thorough and proactive in addressing ethical concerns that comes from this frontier work. But their pioneering work nonetheless raises many open questions: are the xenocortical mice conscious? Do they have any human-like properties of consciousness or cognition? @NitaFarahany and I will be addressing some of these questions in a forthcoming commentary - where we'll also offer an ethically-informed roadmap to guide this important research as it progresses. In this context, "pacing the frontier" really does make sense 😉 Read the Stanford @nature paper here, and buckle up. It's wild. nature.com/articles/s41586-0…

GLiClass gained 100 GitHub ✨ overnight. I suspect the buzz around Jev played a big part, and it’s exciting to see more interest in models that turn natural-language instructions into structured decisions. Back in June 2023, we wrote about our first generalist zero-shot model for classification tasks: blog.knowledgator.com/the-ne… Our long-term vision has been to push these models toward a broader range of tasks, including agentic decision-making. We have also worked on RL agents for path-based reasoning over knowledge graphs. So, how might Jev work, and how does it compare with GLiClass? TypeSafe describes parallel structured outputs and a training method called Reinforcement Learning for Calibrated Decisions (RLCD). Its launch materials do not disclose the exact architecture, parameter count, or complete training recipe. What follows is my hypothesis about the design space: typesafe.ai/blog/introducing… There are several plausible approaches, and we have explored multiple designs for related tasks: • Bi-encoders that align text and label embeddings; • Cross-encoders that jointly process each text–label pair; • GLi*-style models that jointly encode text and multiple labels; • Diffusion-based structured prediction; • Multi-stage encoders with dedicated structured-output heads; • A larger encoder with a small, constrained decoder. These choices can also be combined. The trade-offs depend on input length, the number of questions and choices, interactions between labels, and whether representations can be cached and reused. Given latency and performance characteristics, I would guess the size of the model is a few billion parameters, but it can be easily much larger well optimzied model. My hypothesis would be broad supervised training followed by decision-focused RL, potentially using rewards from multi-step workflows. That could reinforce choices that improve task completion, while calibration would require attention to the quality of the predicted probabilities. For GLiClass, we have already published an RL training approach that adapts PPO to multi-label classification: arxiv.org/pdf/2508.07662 For agentic workflows, state can live in the application and be passed into each model call—as structured context, history, or a compressed summary. Persistent behavior does not require persistent memory inside the model. I also wouldn’t make “encoder or decoder?” the central question. Bidirectional encoders are a strong fit for classification, while causal backbones can also score decisions and benefit from incremental caching. The right choice depends on the workload and training setup. On multimodality, Jev’s published Doom demo uses structured text state rather than images. Its documented interface currently focuses on text and structured state. We remain focused on pushing open-source models toward frontier-level capabilities across more tasks. Stay tuned. Explore GLiClass: github.com/Knowledgator/GLiC…

We have jev at home. The arch is good for RL with state-dependent actionsets, so here is the same arch playing doom and chess (shittily) with two different controllers.

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution

MCDMA Update 🚀 RDMA between METAL and CUDA has been achieved. Started at 600 µs over Ethernet, then 200 µs over USB-C, and now MCDMA’s 4 KiB RDMA medians are Mac WRITE 10.2 µs / READ 7.9 µs and Spark WRITE 4.1 µs / READ 7.1 µs. Still headroom to chase, but it’s out there, so you can have a play and use it in your projects. Next up is an oMLX PR and 25 experiments to see what these machines can do together. github.com/ashhart/MCDMA @sundeep @NVIDIAAI @JensenHuang @nvidia @Apple @johnternus