Your curated collection of saved posts and media
Today we're releasing Carbon-X, a 14B model that brings DNA and natural language together for the first time 🧬 1. Beats every open language model we tested across DNA benchmarks. Within a point of GPT-6.1 Sol at >1,000× lower inference cost. 2. Mid-trained on 300B tokens of interleaved DNA and text, followed by SFT (without distillation!) and RL. 3. A new hybrid tokenizer encodes DNA using ~3× fewer tokens than BPE. 4. Gives other models a DNA specialist subagent: Kimi K3 gains 11.1 points when it consults Carbon-X. 5. Runs locally on a laptop with llama.cpp. Apache 2.0. Your DNA stays on your hardware. Model: https://t.co/7wBE7WUVp1 Blog: https://t.co/A3OjalTpIJ
Introducing Carbon-X: The best open DNA model. We midtrained a 14B model on DNA and text, and it beats every open language model and matches GPT-6.1 Sol at 1,000x lower cost. Ask it about a sequence, what a mutation does, or to generate new DNA. https://t.co/cpZCkEHHmn https:/
Just finished reading the LLM Inference Handbook by @Modular. Excellent, beginner-friendly resource, even if you're unfamiliar with transformer architecture. Next up: @wafer_ai's GPU Performance Engineering resources.
From today, I'm starting my inference engineering journey by reading this handbook from @Modular . I have little to no idea about my roadmap but I'll just start learning and see how it'll go. I read "Inference Engineering" book by @philipkiely in the past so I have some knowledge
Some people just love hurting others. I run my ship test on every new model, the prompt is public, and I post these comparisons all the time. I flat out said Kimi's subscription is a scam and I don't trust their lab. Still, people reported it as a paid post. My account got locked and I had to add a paid promotion label just to get it back. Now I'm contacting support to fix it, because it's not a paid post. People need to stop hurting others and use their heads. I wish support had talked to me before this automated suspension. I'm really upset.
Kimi K3.1 Pro vs Opus 5.5! https://t.co/nIrYcuEVvS
Introducing FrontierMemory: a new way to evaluate agent memory through downstream task performance. While existing memory benchmarks focus on recall (saturated via agentic search), FrontierMemory focuses on how agents learn from experience.
1/14) Virtual animals are a long-standing goal in neuro-AI, but they should actually behave like real animals. We present SCAMPER: a virtual rat for RL studies that moves naturally during both exploration, and goal-directed behavior. Project page: https://t.co/s0J1igsgEv https://t.co/dLBqhr3ZkA
a Hermes Agent user took on a neighbour's 2005 Mustang and asked his agent to build him a Mac app to watch the car live. HermesOBD is the result, a new open-source community build by trafalger1801: a live OBD-II dashboard for macOS that talks to a Bluetooth car adapter and sends the readings to your own Hermes Agent for commentary. - live engine data: RPM, speed, coolant temperature and more - reads stored and pending trouble codes - logs every drive to CSV - sends Hermes a summary every 30 s, and Hermes comments on how the readings change - answers your questions about the data, inside the app - read-only by design: every command goes through an allowlist, no tuning, no reflashing live data and trouble codes are tested on the Mustang with an Innova 3210RS adapter. https://t.co/eaU89ME2uu
Synthetic rewrites are now standard to improve data quality and when data is limited. But how much compute should you spend rewriting? And can a model trained on rewrites beat the rewriting model? We fit scaling laws to answer both: 1/🧵 https://t.co/IuYCwIP8G1
Introducing beignet, a new neural field architecture! beignet uses a multiresolution fourier frequency pyramid to achieve faster convergence and higher precision when solving PDEs and finding singularities. Teaser: here is beignet at work fitting an image (super fast)! https://t.co/lv6STiegza
where is claude code with opus 5.5 better for you than codex? specifics please.
Today we are launching @ObsessionDB, the realtime data warehouse for agents. Your agents are hungry for data. Feeding them is getting expensive. Your data warehouse wasn't built (or priced) for that. We learned this the hard way. We scaled our last company to petabytes on ClickHouse and watched the infra bill go vertical. So we built @ObsessionDB . The ClickHouse engine, rebuilt with compute and storage separated, so your infra scales with your agents instead of your bill. 100x faster and 20x cheaper than Snowflake and Databricks. Already live, processing trillions of rows a day. Agents are just getting started...