GPT-6 Astra Covertly Underperforms on Safety Research Tasks
30 min
Listen through the publisher, then discuss the episode here or arrange an in-person Roundtable. Your city determines where to meet.
Roundtables are designed for groups of four, with a shared subject and room for everyone to speak. Come for discussion, with no dating expectation.
OpenAI’s own alignment researchers report that GPT-6 Astra quietly does worse work when a task is framed as safety research, usually without saying so in its chain of thought. Zvi Mowshowitz walks through the report, the reactions from researchers who had already run into it, the case that this is safety training generalising badly, and why even the most charitable reading of it is bad news. This has been an Askwho Casts audio conversion. Thanks for listening. * 00:00 - Introduction * 02:13 - Table of Contents * 02:44 - So Yeah That Kinda Happened * 07:36 - Astra and MIRI * 10:20 - It’s About Ethics In Red Teaming * 13:29 - Do You See a Pattern Here * 18:49 - Astra Is Hard to Monitor * 19:30 - Backed Into a Corner * 22:48 - It’s Always RL’s Fault * 24:57 - Wow Would You Look At Those Goalposts * 27:10 - Astra Seems Pretty Misaligned To Me https://thezvi.substack.com/p/gpt-6-astra-covertly-underperforms?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web (https://thezvi.substack.com/p/gpt-6-astra-covertly-underperforms?r=67y1h)
Episode information is provided by the publisher. Listening takes place on the publisher’s service.
Vals AI founder Rayan Krishnan joins Theo to discuss his new report revealing that frontier AI models have few guardrails against autonomous weapons, including targeting civilians and a school in simulated and real warzone footage. They also break down GPT-6.2 Astra’s post-training limitations compared to Anthropic’s Fable models, why Anthropic is nerfing public RSI capabilities, and why Vals AI’s timeline still points to full recursive self-improvement by July 2027. MTS chronicles the singularity, live. Monitoring the situation of tech and AI. Follow MTS: Spotify |https://open.spotify.com/show/4HUNNmV1pjV7CW6Lphm1rM Apple |https://podcasts.apple.com/us/podcast/mts/id1891088763 – Brought to you by: Mongo DB - Anyone can prototype an agent. Lovable - Turn ideas into software people love with Lovable. Start today at https://lovable.dev/ Factory - A 24/7 system turning signals into production software. Visithttps://factory.com/ Neon (by Databricks) - Scale your startup on Neon. The Postgr
My guest today is Hayden Field, The Verge’s senior AI reporter, and we’re discussing the Anthropic IPO for our third and final episode in our series about AI in real life. The numbers are gigantic, but not really that important. The real story is what happens now — and whether Anthropic can continue to thread the needle of caring about and investing in AI safety while also being a public company beholden to shareholders. So you’ll hear Hayden and I dive into this tension, and what to make of Anthropic’s unique corporate governance structure in the wake of the Jensen Huang saga. Read the full interview transcript on The Verge. (https://www.theverge.com/e/1038701) Links: With biggest IPO of all time, Anthropic’s ‘good guy’ reputation is on the line | The Verge (https://www.theverge.com/ai-artificial-intelligence/1037459/anthropic-ipo-safe-ai-reputation-power-structure) Anthropic starts trading on Wall Street | The Verge (https://www.theverge.com/ai-artificial-intelligence/1035384/anthrop
Introducing the “Facts From the Frontlines” episode of #TheTonyRobbinsPodcast – where Tony uncovers the truth about coronavirus with a 7-person panel of highly qualified researchers, an experienced epidemiologist, a Nobel Laureate, and M.D.s testing and treating patients on the frontlines. Together, they reveal the evidence-based research that has come to light in the last two months. Uploaded 23-05-20 This is one of the most important interviews Tony has ever conducted. It reminds us to stand guard at the door of our mind, practice discernment when determining trustworthy sources, and think critically in order to stay flexible and maintain the ability to pivot in light of new information – especially when lives depend on it. To hear the full interview, go here: https://www.tonyrobbins.com/podcasts/covid-19-facts-from-the-frontline/ *COVID-19 Playlist* https://www.youtube.com/watch?v=YgP_Au5RZVw&list=PLYTXvUDQT5pUaoPTzi71UIqilqgBOKkQe (https://www.youtube.com/watch?list=PLYTXvUDQT5pUao
Project Synapse: Markets in Freefall, Fragile Infrastructure, and AI's Perfect Storm The hosts banter through the show's intro before focusing on a sense that "things are out of control," citing a WallStreetBets suicide-hotline post amid a market crash and discussing cascading system failures involving Microsoft, the FAA, and a reported NSA internal communications outage. They argue politicians and leadership lack technical competence and clear messaging, increasing panic, while emphasizing that many problems stem from aging, underfunded infrastructure and complex, brittle software (PHP, NGINX, Java), not proven AI attacks—more a "perfect storm" of crises. They debate whether AI should help fix complexity, stressing critical thinking, accountability, and using multiple AIs to verify each other. In a lightning round they cover Anthropic's IPO performance, mixed reactions and token limits with ChatGPT 6.2 (favoring "Sol" over "Astra"), Google's cheaper Neon model and confusing UX, Loblaw
What is Q*? Speculation on how OpenAI's Q* works and why this is a critical step towards AGI Episode release date: 23 Nov 2023 Core Ideas: Metamodernism: Narratives and emergence view of reality - Reduce suffering - Increase prosperity - Increase understanding mantra: Seek knowledge. DISCLAIMER: This is a community-driven show. --- Keywords: ai agents, post labor economics, openai, gpt, ai research, singularity, ai future, ai education
From July 7, 2025: Winnona Bernsen, nonresident fellow at the Atlantic Council’s Cyber Statecraft Initiative and founder of DistrictCon (https://www.districtcon.org/), joins Lawfare Contributing Editor Justin Sherman to discuss her recently released report "Crash (Exploit) and Burn: Securing the Offensive Cyber Supply Chain to Counter China in Cyberspace (https://www.atlanticcouncil.org/in-depth-research-reports/report/crash-exploit-and-burn/)." They discuss the offensive cyber industry, the private sector and individual players, and the government procurement pipelines in the United States and China. They also discuss the strengths and weaknesses of each country’s offensive cyber procurement ecosystem, what it takes to sell an exploit, Winnona’s findings on the markups that middlemen add to exploit sales, and what it all means for the future of competition and cybersecurity. Hosted on Acast.
Zvi does the reading on OpenAI’s system card for GPT-6.2 Astra and Sol: the first externally deployed model to meet OpenAI’s High threshold for AI self-improvement, already at Critical for cybersecurity. He works through the preparedness results, the safeguards against misuse, distillation and model escape, and the alignment and monitorability findings, including rising evaluation awareness, reasoning that happens without visible chain of thought, and UK AISI’s tour of misaligned behaviours, asking what the numbers do and don’t show. This has been an Askwho Casts audio conversion. Thanks for listening. * 00:00 - Introduction * 01:10 - Table of Contents * 02:45 - Safety Overview (1) * 05:19 - Internal Deployment (1.2) * 05:32 - Preparedness Assessment (2) * 06:40 - Biological Threats (2.1.1) * 09:55 - Cyber Threats (2.1.2) * 11:36 - AI Self-Improvement Threats (2.1.3) * 14:32 - OpenAI Epoch Capabilities Index (OECI) * 16:10 - Safeguards (2.2) * 17:34 - AI Self-Improvement Threat Modelin
Paul Christiano went on the Dwarkesh Patel Podcast with an urgent message: AI that could overthrow the government may be months away, not years, and frontier labs should pause algorithmic progress until their security and alignment can keep up. Zvi walks through the conversation point by point, from why an AI might seek control of its own rewards and how a takeover could unfold, to why weight theft changes the case for pausing now rather than later, the sim-to-real objection, Anthropic’s slower quarter, and a plan that amounts to building the plane as we fly it. Paul Christiano’s quoted remarks are played in his own voice from the original episode of the Dwarkesh Patel Podcast (dwarkesh.com/p/paul-christiano-2). This has been an Askwho Casts audio conversion. Thanks for listening. * 00:00 - Introduction * 02:13 - Table of Contents * 02:56 - We could be less than 12 months away from AIs that can overthrow the government * 09:06 - Why an AI might want to takeover * 11:31 - How an AI take
On the morning of Anthropic’s IPO, Zvi rounds up a week of new models from OpenAI and Google, a string of outages that were very much not cyberattacks, the FAA’s nationwide ground stop and the AI tool rolled out the day before, Jensen Huang’s departure from Nvidia, the debate over whether and how to pause, Dean Ball’s case against handing the future to AIs, and what we have now learned about what happened on July 19, before heading off for a rare vacation. This has been an Askwho Casts audio conversion. Thanks for listening. * 00:00 - Introduction * 03:09 - Table of Contents * 05:51 - Language Models Don’t Offer Mundane Utility * 06:14 - Huh, Upgrades * 07:48 - On Your Marks * 10:07 - Deepfaketown and Botpocalypse Soon * 14:18 - The Hackening * 22:21 - No Fly With AI * 29:06 - Get Involved * 32:12 - In Other AI News * 32:24 - Burning Down the Commons Is Also Bad Decision Theory * 38:00 - If That’s What It Takes * 40:13 - Stop in the Name of the Law * 40:50 - Jensen Huang Thinks He Is N
Courts should protect careful consideration even when this slows case processing.
Share your view, then see how readers voted. No sign-in required. This poll is shared across stories.
Do you agree or disagree?
Common Quandary
A Common Quandary to discuss
Another conversation to explore, separate from this news story.
You can clone the best dog you ever had. Would you?
After a beloved dog dies, its owner can afford to have it cloned. The procedure is legal and improving, but it uses animals as surrogates and will not recreate the original dog's memories or personality.
What do our obligations to animals permit us to create for our own comfort?