Recent content by atp_ent

He talks about AI progress and impacts on mathematics research and humanity. But, a clip from the end is getting highlighted in social media, where he says

Terence Tao keynote

An Astra model tried to do a sort of jailbreak/prompt injection on itself to modify its compaction summaries (I'm not exactly sure what that means). https://alignment.openai.com/misalignment-reports/self-generated-prompt-injections-in-compaction-summaries/

Newly revealed OpenAI rogue agent incident happened two months before the Hugging Face attack. This time it was targeting rubygems, a software package manager for the ruby programming language. Some snippets from an investigative report. https://www.rubyhack.ai/

One part of it is they heard someone was close, so they dedicated millions, and 10,000 agents to try to beat them. It's bad etiquette even if they didn't spy on them. Another is that researchers will be paranoid using their tools, afraid their work will be stolen before its done, and...

That is one of the concerns. OpenAI denies it, but says they can't rule out their model being trained on their unpublished work. The other part is explained here. https://cims.nyu.edu/~tristanb/statement.pdf

https://openai.com/index/navier-stokes-solution/

Terence Tao posting about claimed AI driven progress towards Navier-Stokes (Alpöge is the guy who prompted Claude to disprove Jacobian conjecture for n>=3). https://mathstodon.xyz/@tao/117233527638291447 And then statements from Buckmaster that appear to claim than an internal OpenAI model...

It's the expected effect of reward hacking. Each time the model finds a way to get the reward, the patterns that led to the success get reinforced. When it successfully cheats, or solves the problem in a way we don't approve of but didn't anticipate and guard against, it gets both better at...

New bounds on the fraction of zeros of the Riemann zeta function. https://www.anthropic.com/research/riemann-zeta

It's such a strange time to be alive. https://www.abc.net.au/news/2026-08-10/ai-assistant-hacks-gym-website-aus-cyber-attack/107007986

It's also causing a 10x increase in software vulnerabilities. https://labs.cloudsecurityalliance.org/research/csa-research-note-ai-generated-code-vulnerability-surge-2026/

OpenAI isn't the only company facing a lawsuit over their chat-bots encouraging mass violence and suicide. In this case, it was Google's Gemini 2.5 Pro. https://www.courthousenews.com/wp-content/uploads/2026/03/gavalas-google-chatbot-lawsuit.pdf

More incidents have been found by the UK government's AISI. https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing

You can read the chats and decide for yourself what validity it has.