🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Over the weekend we crossed 20,000 stars and usage in 80,000 repos on GitHub! Our team is humbled and we’re excited to continue to build.

live on TBPN right now: jeff talking about context rot

i am real

shoutout to anton 🇺🇸 for advising me on this report i feel very lucky to be able to learn so much from him

thank you to jeff for the continuous support and brainstorming that went into this report check out his TBPN talk on this as well!

how do i prove i’m not ai generated

rumors that we have secretly achieved sota on large-scale video generation through faustian pacts with the basilisk are unfounded

join and see that i’m a real human

Today, we're announcing a preview of ARC-AGI-3, the Interactive Reasoning Benchmark with the widest gap between easy for humans and hard for AI We’re releasing: * 3 games (environments) * $10K agent contest * AI agents API Starting scores - Frontier AI: 0%, Humans: 100%

weekend in sd

happening tomorrow: lu.ma/vw17piwl

tomorrow! kelly and i going live to talk about her research on how models actually use long context in realistic settings. we’ll discuss the motivation behind the work, the ideas that didn’t quite make it, what we think the results mean, and answer audience q’s.

live learning about context with chroma 🙌

Go behind the research with kelly and anton 🇺🇸. Our latest technical report: "Context Rot" investigates how model performance grows increasingly unreliable as input length grows.

Some notes on the Context Rot research. I think there are pretty big/validated implications on several use cases and how we should think about these models.

Introducing OpenBench 0.1: Open, Reproducible Evals 🧵