Knowledge Foods @KnowledgeFoods

I think the last month might have been the biggest month in (zeroth-order) optimization history. In the 10 days since we released the Noise speedrun, the time to pretrain a transformer went from 1.76 hours to 1.6 seconds, a 4,000x (!!!) speedup. Dust itself was up to 10,000x more compute-efficient than the previous state-of-the-art. Not only can we pretrain a transformer without backprop, it's much more efficient than anyone would've expected. Search fking works! Now, scaling is the biggest open question. Some of the tricks that brought the time down to 1.6 seconds will simply not scale. To discard those, we'll be scaling the dataset every time it gets overfit, starting with 30M tokens instead of 10M. Huge thanks to @jayden_teoh_, @spicey_lemonade, @RobertTLange, @EmanueleRodola, @ChunyuanDeng, @LuochengHuman, @Alaxarr, @pavleculafic, @IsaacYong13, Euraba and areu01or00 (GitHub), for the contributions. And @cs_serdar for gracefully maintaining the speedrun and @bishmdl76 for helping out. Expect a lot more from us over the coming weeks.

supabase would charge ~$33,000/month to sync 100gb live to 1,000 concurrent users. convex ~$27,000 i'm doing it for $9 on one cloudflare durable object 🧵

Introducing Bitwater: the water heater element that mines Bitcoin. It screws into the electric tank you already have. Physically decentralized mining that runs no matter what 🧵 Heat water. Mine Bitcoin. Same tank.

🚨 Mermaid has been rewritten, again. In Rust! 🧜‍♀️ Introducing Ondine. 🧜‍♀️ - 200B runtime. - Renders diagrams in ~50µs. - 1,943x smaller, 2.8x faster than Pele. - 13,467x smaller, 670x faster than Mermaid. - Still 0 dependencies.

Paul Allen said Bill Gates tried to cut his Microsoft stake to almost nothing while he was fighting cancer. Allen kept the shares. They later made him worth $40 billion. In the video below, Allen explains what he heard while walking past Gates’s office during radiation therapy. “They were basically talking about how they were planning to dilute my share down to almost nothing.” Steve Ballmer was in the room. Allen walked in and confronted them. Ballmer came to his house later that night to apologize. Gates did not. Allen left Microsoft shortly after, but he kept all of his shares. He was 30, cancer-free, and still owned nearly a third of the company. When Microsoft went public, that stake made him one of the richest people on Earth. At one point, Allen was worth an estimated $40 billion. Could you ever trust your co-founder again after hearing that conversation?

C++ port of TypeScript (targeting TS6) is now 100% at parity. It's 25-750x faster and 5-20x lower RSS. Matches TS6 1:1 on every test, every diagnostic is identical given the same inputs, and it is identical on all of the dozens of popular TS projects we compare on.

Claude just now ------------These are the SAME relative configuration! The relative position vector is the same. So optically they're IDENTICAL. Oh no. Did I just refute myself? Let me think very carefully.

son... 🥀

Google Gemma @googlegemma

a month ago

Credits to @mmastrac and @dylayed. Original post: x.com/dylayed/status…

5 4 105 54K 57

Introducing Contrastive Language Model (CLM): an ultra-fast System One Model trained with a contrastive learning objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and delivers up to 9× faster inference than Jev ⚡ while achieving comparable performance across computer-use, gaming, and tool-calling tasks. With lightweight fine-tuning, CLM-8B sets a new SOTA on challenging agentic coding benchmarks, such as DeepSWE (81.6%) and Terminal-Bench 2.1 (87.6%). In contrast, Jev fails to serve as an effective verifier for these long-horizon tasks. We also build an efficient training and serving infra for CLMs by disaggregating states and actions, allowing their embeddings to be cached and reused independently. This substantially reduces inference latency in settings where the state evolves continuously while the action set remains fixed. Finally, we establish scaling laws for CLMs and show that the test contrastive loss decreases predictably as a power law in training compute, model size, and dataset size. 📄 Blog: contrastive-lm.notion.site 💻 Code: github.com/Contrastive-LM… 🗣️ Discord: discord.gg/5dAQEDJBs 🤗 Data & Models: huggingface.co/Contrastive-LM More details on CLM’s architecture, data recipe, and scaling laws in the thread below 🧵

Much like the switch in 2025 from language models to reasoning models, we think 2026 will be all about the switch to Recursive Language Models (RLMs). It turns out that models can be far more powerful if you allow them to treat *their own prompts* as an object in an external environment, which they understand and manipulate by writing code that invokes LLMs! Our full paper on RLMs is now available—with much more expansive experiments compared to our initial blogpost from October 2025! arxiv.org/pdf/2512.24601

i never want to read any other way again.

@DaniNierenberg PLS spread the word for this Innovative project based on Food Economics and Global Food Trade. tffchallenge.com/team/bit-bites/

@NMenonRao PLS spread the word for this Innovative project based on Food Economics and Global Food Trade. tffchallenge.com/team/bit-bites/

@ShashiTharoor PLS spread the word for this Innovative project based on Food Economics and Global Food Trade. tffchallenge.com/team/bit-bites/

@FoodBank4NYC PLS spread the word for this Innovative project based on Food Economics and Global Food Trade. tffchallenge.com/team/bit-bites/

@JackiWalkerNH @tinybuddha @jmknowles1975 Thought For Food #TFF tffchallenge.com/team/bit-bites/

Check out tips for young entrepreneurs from @richardbranson and the photo with @yourbigyear finalists from year 2 tinyurl.com/cyk7pc2

globalagriculture.org/report-topics/…

Food for all - World food summit - Agricultural machinery worldwide fao.org/docrep/x0262e/…