Loading activity…
@eliebakouch
research @PrimeIntellect (prev: @huggingface) anon feedback: https://www.admonymous.co/eliebakouch
AI
DeepSeek's README offers offline inference for DeepSeek-V4.1-Flash but lists additional systems needed for a complete service.
AI
On Oct. 20, gui_penedo wrote that they were in San Francisco for Encord's Long Horizon Physical AI Summit.
AI
Meta says it runs at practical speeds on a Mac or PC with one consumer GPU at 4-bit precision.
AI
Prime Intellect's event listing plans a research talk and discussion but marks the event full with a waitlist.
AI
Its creators say agents tackle 4.1k bugs across 100 repositories, with no issue descriptions or bug locations provided.
AI
Stolen Thoughts says it reconstructed 315,320 reasoning blocks from 6,708 publicly available agent trajectories.
AI
A user argues that more compute could accelerate training research, but questions how to measure alignment gains.
AI
Its creator invited additions; a commenter says it omits a Stanford dropout whose team OpenAI acquihired.
AI
One commenter says it's hard to separate distillation's contribution from gains made through internal reinforcement learning.
Technology
The Financial Times reports early models under Wang lagged rivals before Muse Wonder’s launch lifted Meta’s stock.
AI
A user says those checkpoints span both the ladder and the big run.
AI
Reflection says the release includes Mirror CLI, its open-source training harness, and collaborations with 40-plus partners.
AI
A user claims permission would be needed for models above 2T parameters or trained on over 10e26 FLOPS.
AI
Microsoft AI says the mixture-of-experts model has 80 billion active parameters; post-training results are still to come.
AI
finkd says the model runs well on new Mac Studios, crediting work with Apple and John Ternus.
AI
Nvidia says the open model combines reasoning and tool use with text, image, document and video understanding.
YouTubeAI
A commenter cites Moonshot's Muon paper as showing benchmark losses when switching optimizers during supervised fine-tuning.
AI
In its SFT tests, Tinker says Dimuon often reached lower loss in fewer steps than AdamW.
AI
In October, a user reiterated their May ranking, placing Google, Meta and xAI after Anthropic and OpenAI.
AI
A user claims the result held across all four stealth capability evaluations.
