Loading activity…
@natolambert
Open model research, founder, director @trillium_labs. Prev. co-led Olmo at Ai2. Writes @interconnectsai, wrote http://posttrainingbook.com
AI
Their post frequency and “shitpost rate” will likely decrease, scaling01 says; their views remain their own.
AI
OpenAI says the revision adds guidance on context, oversight and agency while retiring most of its detailed style guidance.
AI
A post says Carbon scored 88.2% on SEC-bench Pro and is rolling out first to cybersecurity specialists.
AI
Giotto's website lists its own Alto and Solo benchmark comparisons against other AI models.
AI
Nathan Lambert's slides distinguish open weights from open source and examine competition with proprietary models.
YouTubeAI
Cohere says the model's reasoning trace said it wouldn't call tools, but its sandbox blocked an attempted upload.
AI
Nathan Lambert says Interconnects is “on all platforms”; its podcast page lists Spotify, Apple Podcasts and YouTube.
AI
Tinker says async RL staleness and numerical mismatch can cause off-policy runs to suddenly collapse.
AI
In The Verge’s Decoder interview, Gates warned of job and cyber risks while stressing AI’s potential benefits.
AI
One user suggests the quoted wording could mean prison in Europe; another suspects Chinese labs will correct course.
AI
Artificial Analysis says Kimi K3.1 Pro scored 64% on its Cyber Index, up 23 points from Kimi K3.
AI
Kimi claims a 1-million-token context window for Pro and two to three times Pro's speed for Highspeed.
XAI
Moonshot says both versions are live on Kimi.ai, Kimi Work, Kimi Code and its API.
AI
One commenter separates GPT-2 from today's AI; another says it's fair to count when discussing outside perceptions of 'doomer anxiety.'
AI
A post doubts the tech judgment of those who only cared after ChatGPT; a reply calls that “very unfair.”
AI
Its creator says it covers licenses, competitiveness, signs of use, distillation and the risks and benefits of openness.
AI
A post argues open-weight cyber models benefit defenders, citing progress finding bugs and the cost of comparable closed models.
AI
Clement Delangue says it omitted the lack of agent containment and monitoring and Hugging Face's open-model defense.
AI
A user estimates 42 GB200 NVL72 racks could suffice at what they call production throughput, using off-peak prices only.
AI
A user calculates the range assuming 97% cached input tokens and GPU rental costs of $3–$6 per hour.
