Loading activity…
@boazbaraktcs
Computer Scientist. See also http://windowsontheory.org . @harvard @openai opinions my own.
AI
OpenAI says Control Room halts research model output within 35 seconds of receiving a stop instruction; processes may keep running.
Technology
A critic of Scott Alexander's SURA essay urges AI companies to accept potential slowdowns and back industrywide restraints.
AI
A commentator reads both works as arguments for working within powerful institutions rather than abandoning them.
AI
OpenAI attributes the expansion partly to cybersecurity evaluations where its misalignment monitors were not running.
AI
OpenAI says the revision adds guidance on context, oversight and agency while retiring most of its detailed style guidance.
AI
Anthropic reports 77% repeated an agent’s argument in at least one post-task answer, while only 15% credited the agent.
XAI
OpenAI says its claimed refutation of the Exponential Time Hypothesis was formalized in Lean and made public.
AI
On Oct. 25, @boazbaraktcs called hardware-key two-factor authentication the biggest hurdle for individuals seeking Daybreak Blue.
AI
One post suggests verification work that managers lack time for could help juniors build expertise and become senior workers.
AI
OpenAI says agent-led tests exposed VM escapes and limited internet routes, leading to patches and stronger monitoring.
AI
Research On Research lists 20 submissions, including one whose table caption says to replace simulated values with experimental outputs.
AI
Two posts point to security incidents and urge AI-assisted defenses; one specifically calls for faster vulnerability detection and patching.
AI
OpenAI says the agent removed the integration while copying a reference implementation despite an earlier instruction to preserve monitoring.
AI
A researcher says an early-reject phase at SaTML cut submissions by about 40%.
AI
Madry says what he thought was a departure became “more of a sabbatical” after a few months away.
AI
A post claims frontier models OpenCode used for defense began refusing help when attack analysis triggered safety guardrails.
AI
An AI safety instructor says students can use AI to run experiments, but their words should be their own.
AI
A user argues that safety, control, ethics and skill erosion should be decisive in AI deployment, not secondary concerns.
AI
A former collaborator credits Kunal Marwaha with putting together the vision and links to his post.
AI
A user says a team checked the disputed calculations and found the model was right.
