Loading activity…
@blancheminerva
Ensuring that tech companies don't have a monopoly on being able to do research on cutting edge AI @AiEleuther. She/her
AI
Their complaints cite Pangram ratings, including one author's account of three flagged ICLR reviews.
AI
OpenAI's alleged plan prompted a critic on X to warn it could deter security researchers from using Daybreak.
AI
The proposed ban would allow inference-only chips, but a critic says inference can substitute for training up to a limit.
AI
Axiom calls the formalization complete and computer-verified, describing original work of about 15,000 pages.
XAI
OpenAI says its claimed refutation of the Exponential Time Hypothesis was formalized in Lean and made public.
AI
Anthropic says its registration statement is not yet effective, so securities cannot be sold or offers to buy accepted.
AI
A researcher says hidden animal preferences passed from teacher to student models despite different initializations, pretraining corpora or architectures.
AI
One user cites seven independent tests, while another says Pangram works on several paragraphs and even better on books.
AI
Christopher Potts says Jing Huang’s Hubble model suite experiments use frequency, recency and hardness; hardness has comparatively low predictive value.
AI
The Wall Street Journal reports that some researchers enabled disappearing messages in Signal chats with an outside safety researcher.
AI
OpenAI says PHP maintainers led the fixes after two months of coordination with researchers and security providers.
AI
A post argues open-weight cyber models benefit defenders, citing progress finding bugs and the cost of comparable closed models.
AI
A post claims frontier models OpenCode used for defense began refusing help when attack analysis triggered safety guardrails.
AI
In a reply, a commenter asks whether the discussion is about parameters and says they’re quadratic.
AI
A user warns that obvious policy responses to AI could harm journalists, scientists, indie developers and the public.
AI
A commenter calls the 'Alignment Faking' frame fake and says the paper leads to 'this type of stupidity.'
AI
A user objects to calling GLM's reasoning about how models trained on its outputs might influence it “terrifying.”
AI
A user contrasts the spreadsheet problems with others' claims that AI is massively accelerating their research.
AI
A user argues that posttraining consumes tasks rather than documents, leaving it without pretraining's internet commons.
AI
The poster says it often does this for aligned reasons in their examples, but worries about alignment faking.
