SignalTech/AI
I struggle to believe that AI R&D is ingenious
High-priority alert based on source significance and keyword relevance.
Context
What happened: A commenter argues that AI models show poor basic judgment in R&D tasks, citing failures to identify important papers, experiments with no information value, and trivial research directions, while noting some success on fast feedback hill-climbing tasks.
Key numbers: Eight hackathon papers scored between 60% and 95% on
Why it matters: The commenter questions whether this bears on current public models speeding up AI R&D or only on fully autonomous R&D, and disputes lab insiders' claims that internal models outperform Astra, noting such claims recur each model generation.
What changed
A commenter argues that AI models show poor basic judgment in R&D tasks, citing failures to identify important papers, experiments with no information value, and trivial research directions, while noting some success on fast feedback hill-climbing tasks.
Why it matters
The commenter questions whether this bears on current public models speeding up AI R&D or only on fully autonomous R&D, and disputes lab insiders' claims that internal models outperform Astra, noting such claims recur each model generation.