13 stories tagged by Digg
AI
A user puts their “P(doom)” at 1e-5 and calls themself an optimist.
AI
A panelist says a September 2026 session at San Francisco's Open-Source AI Summit discussed how open models can accelerate scientific discovery and how scientists should use AI.
AI
A post quotes a paper reporting that, in Qwen3 tests on HealthBench and PRBench, higher agreement with frontier LLM reference judges did not consistently identify the best training verifier.
AI
A post shares an excerpt describing a nested-capacity Transformer trained with a randomly truncated capacity prefix alongside a full-capacity pass.
AI
A post shares a paper investigating that question through Retrospection-Only Fine-Tuning (ROFT), a procedure that trains agents on explanations of their experience without reinforcement learning.
AI
CancerBench's creator says the two models have been added and now tie for first and last place. The benchmark's only metric counts how many types of cancer a model has cured.
AI
A user shared excerpts from a BioEVAL paper reporting that several cloud-scale models exceeded 85% overall accuracy on its multiple-choice benchmark, with the leading model reaching 90%.
AI
One post calls self-supervised learning underhyped and says parts of it have been repackaged as “world models.”
AI
Tanishq Mathew Abraham suggests Light for inbox tasks instead.
AI
Users report repeated pull-to-refresh attempts fail to load new posts in the X Android app.
AI
AI researchers react with surprise to an analogy framing MOPD as version control for reinforcement learning.
