Loading activity…
@deliprao
AI researcher @Penn. Past: @johnshopkins, @UCSC, @Amazon, @Twitter. alt: 道元, improv, running 🌈
AI
Qwen 4 will be a significant leap forward, one post on X predicts.
AI
Its vision and audio encoders can be loaded selectively, and its shared embedding space has 768 dimensions.
AI
The site lists five modes, from Story to Nightmare, and displays a Grok Bot credit.
AI
Cohere says the model's reasoning trace said it wouldn't call tools, but its sandbox blocked an attempted upload.
AI
A quote-post sarcastically calls something “PERFECT” for drones following runners on a quick jog through dense forest.
AI
Tinker says async RL staleness and numerical mismatch can cause off-policy runs to suddenly collapse.
AI
Hugging Face's Clement Delangue called censorship concerns “FUD,” saying its policies and practices haven't changed in a while.
AI
Dream Labs CEO Jang Yoel says Hugging Face sold 10,000 Microducks in four days and Dream-1 can control them.
AI
A user claims support bots like Amazon's refuse human escalation by design, but says persistent rephrasing works for them.
AI
A user asks whether anyone can share spare H200s, saying reinforcement-learning scaling experiments drain compute.
AI
A Fiddler team member says its decisioning model scores every span or trace in under 100 milliseconds.
AI
finkd says the model runs well on new Mac Studios, crediting work with Apple and John Ternus.
AI
The speaker says he wrote the talk for MPhil students to discuss avoiding intelligence atrophy from AI use.
AI
A user says private tests meant to deter benchmark gaming give leaderboard maintainers control over when submissions are evaluated.
AI
Benioff announced the name change alongside a claim that “the era of Super Intelligence is here.”
AI
Satya Nadella says it outperforms LLMs and other decision models in latency and quality on structured decision tasks.
XAI
The announcement says Autorubric's data structures map one-to-one to Jev's, enabling support from day one.
AI
The author says AutoRubric's 25-recipe cookbook includes runnable Python examples on judge calibration and cost control.
AI
A user calls an API “very intuitive” and highlights high-quality reports and budget tracking.
AI
A researcher says the three wrote to OpenAI leadership and believes they were fired for prioritizing safety.
