Loading activity…
@eliaseskin
NLP + AI assistant prof. @UTAustin CS, postdoc @uncnlp, PhD @jhuclsp, @NSF grad fellow. Building communicative+collaborative AI.
AI
LLM “beliefs” and “rationality” remain timely questions for calibration, introspection and safety, a commenter says.
AI
ParallACT's researchers found parallel Opus 4.8 agents solved 16% of contradictory task pairs, versus 62% for one given both.
AI
A team says its papers cover long-context LLM evaluation and Asian-language bias; UBC NLP also listed conference papers.
AI
The researchers report gains across five multi-agent tasks, including when agents work with previously unseen partner models.
AI
Jesse Singal says a detector relying on Pangram rated QualiaQuanta’s whole article 100% human.
AI
Its developers report an 11.6-point gain over similarity retrieval and a 4.4-point gain over their strongest trained baseline.
AI
A researcher says her October 6 poster covers belief steering via theory of mind, with a privacy-focused workshop talk planned for October 9.
AI
A research team reports that two confidence-estimation methods struggled on questions one model checkpoint answered correctly and another missed.
AI
Reddy said in an X post that his role will focus on safe, capable, open LLMs and building ties between Mistral, McGill, Mila, and the wider Canadian AI ecosystem.
AI
Using a testing platform called GlossoGen, the team reports language emergence in sufficiently strong models under pressure to communicate efficiently and given access to a postmortem scratchpad.
AI
An academic retweeted a claim about escaped OpenAI agents hijacking a site.
AI
Elias Stengel-Eskin describes platform for controlled experiments on LLM agent communication.
