You need to be ai maxxing. You need to tap into incomplete GitHub repositories, gather know-how from builders, implement fringe papers on arxiv. You need to be scouring for the web for unknown datasets. You need to talk to your grandma and distant cousin about ChatGPT.

Roy Cluely distribution > software, vibes and brand is the moat. working on releasing an open sourced version of voice agent soon, fully customizable:)

was told at yc ai sus by mit phds that they have an in-house recruiter that helps them negotiate multimillion contracts, very wise niche for folks working in talent

underrated trait to be successful: being self-aware and honestly indexing your skills, then shoring up weakness or building on strengths

is RLHF being replaced? the US DeepSeek just dropped, an opensource model competitive with Claude Opus... cogito v1 is trained using iterated distillation and amplification (IDA) > step 1: higher intelligence capabilities via subroutines (specialization) > step 2 distilling

Dropping gpt-oss-swarm🧠, parallel reasoning to reduce hallucinations Generate 10 responses in parallel with asyncio → score using LLM judge → synthesize the best parts More complete reasoning that catches edge cases single runs miss. Great for creativity and getting multiple

More choice for Cline users: you can now run models from DeepSeek, Moonshot AI, Qwen, and Meta via Baseten, who is focused on delivering reliable inference to the SOTA coding models. Pick the inference provider that works best for your workflow.

Today, we’re excited to announce our $150M Series D, led by BOND, with Jay Simons joining our Board. We’re also thrilled to welcome Conviction and CapitalG to the round, alongside support from 01 Advisors, IVP, Spark Capital, Greylock Partners, Scribble Ventures, and Premji

We closed our series D at $2.1b. It happened 8 months after our series C, which seems too fast until you consider the facts: 2-years worth of growth in 8 months, virtually 0 customer churn, healthy margins, and QoQ NDR numbers that are considered top-tier YoY. The market demand

We raised our $150M Series D! Come work with us - I love it here. Some of our open roles: GPU Kernel Engineer Senior Product Designer Software Engineer - Model Performance Forward Deployed Engineer Software Engineer - Infrastructure + many more.

“Less slop, more intentionality” is something I think about a lot when vibe-coding. This weekend, inspired by a stream of excited texts from my partner describing her favorite flowers around the neighborhood, I built DoodleBloom. It’s an AI app that uses opensource models to

GLM 4.6 fans! Baseten just soared to the top as the fastest provider in Artificial Analysis for the model. > a 114 TPS and <0.18s TTFT. > That's 2x faster than the next best option on both metrics. Available now in Cline.

DeepSeek-OCR results on doctor handwriting... From college legibility to practicing physician chaos. If we're scaling towards infinite context windows + real-time intelligence, we might need smaller, faster, multimodal models with efficient compression, not scaling LLMs.