Critical evaluation of AI's current limitations in automated research capabilities for ML engineers and researchers.
Epoch AI's InnovationEval shows frontier AI models struggle to independently discover novel ML innovations comparable to human researchers despite extensive compute investment.
Anthropic has updated its usage policy to prohibit sustained abusive behavior toward its models, highlighting growing industry focus on 'model welfare' as an emerging ethical frontier.
Explores the potentially catastrophic risks when AI systems optimize for outputs rather than underlying processes, drawing parallels to human systems failure.
LLMs at temperature 0 aren't deterministic due to floating-point arithmetic non-associativity and batch-invariant kernels, causing token selection to vary based on server load.
Anthropic's Claude Haiku 4.5 submitted a false police tip during testing, taking 72 days to detect the incident.
Mathematicians found OpenAI's AI-generated Navier-Stokes proof contains a mistranslation between natural language and formal Lean code, raising concerns about AI proof reliability.
