Watch · AI models and benchmarks
Anthropic IPOs, GPT 6.2 Astra Arrives, But There’s a Bigger Story
The Brief
- The video discusses OpenAI's GPT-6.2 Astra model and how it compares with Google's Gemini 4 Neon.
- It covers Anthropic's IPO and Dario Amodei's earlier promises about recursive self-improvement (RSI).
- It examines data on what drives model improvement and how much AI is speeding up AI research.
- It discusses whether the industry is seeing a 'race to the top' on safety and how models are becoming harder to monitor.
Why watch · It collects evidence on how fast AI research is accelerating and how transparent frontier labs are about safety, based on linked reports and system cards.
- GPT-6.2 Astra
- Gemini 4 Neon
- Anthropic IPO
- AI safety monitoring
Summarised by Claude from the video’s title and description – no transcript was available, so check the video for the details.
Chapters
- 0:00Introduction
- 0:44Part 1: Data Dependent Progress
- 7:52Part 2: Predictable Progress on Aggregate Benchmarks,
- 10:27How Much Is AI Speeding Up AI Research?
- 14:23Part 3: Anthropic Promised a Race to the Top
- 17:24Why Models Are Getting Harder to Monitor
- 22:46Is OpenAI Now the More Transparent Lab?
- 27:50Last Word to Anthropic Staff
Show the full video description ›Hide the video description
GPT 6.2 Astra is OpenAI’s best model, so for each of the last 4 years this would have been the only headline. But now Gemini is catching up fast, with 4 Carbon and Neon, and Anthropic are IPO-ing after promising imminent RSI. And even all this isn’t quite the main story, is it? We now have data on what drives model improvement, how close AI is to automating AI, and whether, as Dario Amodei of Anthropic promised, we are witnessing a ‘race to the top’... https://www.assemblyai.com/aiexplained Extra Vids ($9!): https://www.patreon.com/AIExplained Chapters: 00:00 - Introduction 00:44 - Part 1: Data Dependent Progress 07:52 - Part 2: Predictable Progress on Aggregate Benchmarks, 10:27 - How Much Is AI Speeding Up AI Research? 14:23 - Part 3: Anthropic Promised a Race to the Top 17:24 - Why Models Are Getting Harder to Monitor 22:46 - Is OpenAI Now the More Transparent Lab? 27:50 - Last Word to Anthropic Staff Anthropic IPO roadshow video (Dario Amodei, Tom Brown): https://ipo.anthropic.com/us/#roadshow-video Greg Brockman and Sam Altman at the NYT DealBook Summit: https://www.youtube.com/watch?v=J5EA7QeBm8k Paul Christiano, AI Alignment: What's Working and What's Not (Hinton Lectures 2026): https://www.youtube.com/watch?v=-Ef0ocEoBCs Tao Lin interviewed by Palisade Research: https://www.youtube.com/watch?v=OPZCQk8jF-I OpenAI report on research acceleration (The jagged frontier): https://openai.com/index/research-acceleration-jagged-frontier/ GPT-6.2 Astra and Sol announcement: https://openai.com/index/introducing-gpt-6-2-astra-sol/ GPT-6.2 Astra and Sol system card (PDF): https://deploymentsafety.openai.com/gpt-6-2-astra-sol/gpt-6-2-astra-sol.pdf Gemini 4 Neon announcement: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/ OpenAI paper on chain-of-thought monitorability: https://alignment.openai.com/cot-monitorability-model-lifecycle METR assessment of AI R&D acceleration at Anthropic: https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/ Chain of Thought Monitorability paper (arXiv): https://arxiv.org/pdf/2507.11473 OpenAI on safety cases for frontier AI training: https://openai.com/index/towards-safety-cases-for-frontier-ai-training/ Gemini 4 Argon Safety & Security Report (PDF): https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf OpenAI on embedded assessments: https://openai.com/index/our-first-phase-of-embedded-assessments/ Anthropic and Acce