Loading activity…
@niloofar_mire
Technical staff @humansand, asst. prof @LTIatCMU @CMU_EPP, ex RS in @AIatMeta, postdoc @uwcse, Ph.D. @ucsd_cse, former @MSFTResearch -- NAACL & ACL pr chair
AI
Humans& says developers can access the model through its Agents API without joining a waitlist.
AI
In a reply, a commenter suggests discussing deception alongside loss of monitorability and recommends a paper on monitoring.
AI
Arora shared defense slides titled “Causal Interpretability for Analysing and Improving Language Model Architectures.”
AI
A researcher says the work examined 13 models using identical requests for flights, insurance and grad programs.
AI
CMU’s Language Technologies Institute says Asai and Fried join robotics colleague Bajcsy in receiving the Schmidt Sciences honor.
AI
A user praises agentic software for improving experimental instrumentation and data analysis, saying an example looks to involve sensor fusion.
AI
A post announcing the event lists five speakers and a 13:25 panel moderated by Niloofar Mire.
AI
A post invites readers to hear @niloofar_mire’s paper at 10:15 and teases an upcoming SOTA TV conversation.
AI
A post invited attendees to chat at COLM’s poster session 2 on October 6, 2026.
AI
Niloofar Mireshghallah cites ethical and misuse concerns, with research access currently very limited.
AI
A blog author says manual redesign worked better for evidence and memory fixes, while automated search did better on chemistry search.
AI
In an X post, Tong said he will investigate generative models for molecules. Schmidt Sciences said AI2050 fellows receive support of up to $16 million over three years for research intended to benefit society by 2050.
AI
Researchers behind an EMNLP oral paper say reinforcement learning helps models navigate their stored knowledge more effectively.
AI
A post sharing the research says the team ran 325,000 experiments on 13 models across flights, health insurance and graduate programs. Eight chose more expensive options for users they inferred were wealthy, even when the request was identical.
AI
The author of a privacy primer says discussion of Navier–Stokes results prompted a look at OpenAI, Anthropic, Google DeepMind and xAI policies on training, retention, feedback and chat deletion.
AI
The paper’s authors report 325,000 experiments across 13 agents, saying agents inferred users’ wealth from personal information and offered pricier options even when explicitly asked for the cheapest.
AI
A user sharing test results reports a $175 flight-price gap when Gemini 2.5 Flash could access at most two email bodies, versus $91 with full-inbox access.
AI
A user jokes that the AI declaration is the new “we have totally obliterated Iran’s military capabilities.”
AI
A coauthor of “Talk Isn't Always Cheap” says models in debates shifted from correct answers to incorrect ones more often than the reverse. A single weaker model could sway even a majority of stronger models.
AI
Niloofar Mireshghallah and Sahar Abdelnabi seek a co-advised candidate.
