#222 - Sora 2, Sonnet 4.5, Vibes, Thinking Machines
Last Week in AI
00:00
Evolution of Concepts in Model Pre-Training
Summary of mechanistic interpretability work tracking feature emergence during pretraining and implications for steering and safety.
Transcript
Play full episode
Transcript
Episode notes
Our 222st episode with a summary and discussion of last week's big AI news!
Recorded on 10/03/2025
Hosted by Andrey Kurenkov and co-hosted by Jon Krohn
Feel free to email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
In this episode:
- OpenAI introduced several new features, including SOA 2 for text-to-video generation, Claude Sonnet 4.5 for coding and agentic tasks, and the pulse feature for personalized morning briefs.
- Meta launched a new AI video creation feature called Vibes in its Meta AI app and on meta.ai, facing mixed reactions from the public regarding AI-generated content.
- California's SB 53, the Transparency in Frontier AI Act, has become law, requiring large AI companies to disclose safety and security processes, while SB 942 mandates AI detection tools for user-generated content.
- AI regulations and industry dynamics, including battles over intellectual property, startup funding, and the integration of AI into everyday tools and services like Microsoft's AI agents for Word, Excel, and PowerPoint.
In this episode:
- (00:00:10) Intro / Banter
- (00:03:08) News Preview
- (00:03:56) Response to listener comments
- Tools & Apps
- (00:04:51) ChatGPT parent company OpenAI announces Sora 2 with AI video app
- (00:11:35) Anthropic releases Claude Sonnet 4.5 in latest bid for AI agents and coding supremacy | The Verge
- (00:22:25) Meta launches 'Vibes,' a short-form video feed of AI slop | TechCrunch
- (00:26:42) OpenAI launches ChatGPT Pulse to proactively write you morning briefs | TechCrunch
- (00:33:44) OpenAI rolls out safety routing system, parental controls on ChatGPT | TechCrunch
- (00:35:53) The Latest Gemini 2.5 Flash-Lite Preview is Now the Fastest Proprietary Model (External Tests) and 50% Fewer Output Tokens - MarkTechPost
- (00:39:54) Microsoft just added AI agents to Word, Excel, and PowerPoint - how to use them | ZDNET
- Applications & Business
- (00:42:41) OpenAI takes on Google, Amazon with new agentic shopping system | TechCrunch
- (00:46:01) Exclusive: Mira Murati’s Stealth AI Lab Launches Its First Product | WIRED
- (00:49:54) OpenAI is the world's most valuable private company after private stock sale | TechCrunch
- (00:53:07) Elon Musk’s xAI accuses OpenAI of stealing trade secrets in new lawsuit | Technology | The Guardian
- (00:55:40) Former OpenAI and DeepMind researchers raise whopping $300M seed to automate science | TechCrunch
- Projects & Open Source
- (00:58:26) [2509.16941] SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks?
- Research & Advancements
- (01:01:28) [2509.17196] Evolution of Concepts in Language Model Pre-Training
- (01:05:36) [2509.19284] What Characterizes Effective Reasoning? Revisiting Length, Review, and Structure of CoT
- Lighting round
- (01:09:37) [2507.02954] Advanced Financial Reasoning at Scale: A Comprehensive Evaluation of Large Language Models on CFA Level III
- (01:12:03) [2509.24552] Short window attention enables long-term memorization
- Policy & Safety
- (01:18:11) SB 53, the landmark AI transparency bill, is now law in California | The Verge
- (01:24:07) Elon Musk's xAI offers Grok to federal government for 42 cents | TechCrunch
- (01:25:23) Character.AI removes Disney characters from platform after studio issues warning
- (01:28:50) Spotify's Attempt to Fight AI Slop Falls on Its Face
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.
The AI-powered Podcast Player
Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
