Our 244th episode with a summary and discussion of last week's big AI news!
Recorded on 05/08/2026
Hosted by Andrey Kurenkov and Jeremie Harris
Feel free to email us your questions and feedback at [email protected] and/or [email protected]
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
In this episode:
- OpenAI released GPT-5.5 Instant as ChatGPT’s new default model, showing large benchmark gains and crossing a “high” cyber-risk threshold under its preparedness framework, while bio-safety results were mixed.
- OpenAI investigated and patched ChatGPT’s “goblin” obsession, attributing it to reinforcement-learning rewards that over-amplified playful creature metaphors in a nerdy persona that later bled across versions.
- Major industry moves included xAI’s Grok 4.3 price cuts and voice tools, Mistral’s unified Medium 3.5 model and Work mode, and Anthropic’s managed-agent upgrades alongside a surprise SpaceX compute deal and reports of a much higher Anthropic valuation.
- Key policy and security developments covered the Musk–OpenAI trial details, Pentagon AI deployments on classified networks, expanded U.S. government pre-release model reviews, and reports of NSA testing Anthropic’s Mythos on Microsoft software.
Timestamps:
- (00:00:10) Intro / Banter
- (00:01:14) News Preview
- (00:04:39) Response to listener comments
- Tools & Apps
- (00:13:40) OpenAI releases GPT-5.5 Instant, a new default model for ChatGPT | TechCrunch
- (00:18:23) ChatGPT Became So Obsessed With Goblins That OpenAI Had to Intervene
- (00:27:14) xAI launches Grok 4.3 at an aggressively low price and a new, fast, powerful voice cloning suite | VentureBeat
- (00:33:49) Mistral's new flagship Medium 3.5 folds chat, reasoning, and code into one model
- (00:39:28) Anthropic updates Claude Managed Agents with three new features - 9to5Mac
- (00:43:42) ElevenLabs Revamps AI Music Platform as Fan-Focused Service
- Applications & Business
- (00:44:57) A diary, a threat, and a $30 billion stake: What the Musk vs OpenAI trial has actually shown in its first week - The Times of India
- (00:55:28) Anthropic, SpaceX Sign Deal to Boost AI Computing Power for Claude Software - Bloomberg
- (01:01:48) Anthropic in talks with investors to raise funds at $900 billion valuation, higher than OpenAI
- (01:02:37) Anthropic and OpenAI are both launching joint ventures for enterprise AI services | TechCrunch
- (01:06:15) Anthropic and FIS Are Building an AI Agent to Help Banks Police Financial Crimes
- (01:07:02) AMD’s revenue jumps 38 percent from last year as Q1 data center sales hit $5.8 billion. | The Verge
- (01:08:51) Banks seek to offload risk to avoid ‘choking’ on data centre debt
- (01:14:08) DeepSeek could be valued at up to $50 billion in first fundraising, sources say | Reuters
- Projects & Open Source
- (01:16:14) Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
- (01:22:23) OpenAI just open-sourced its data center networking technology
- Policy & Safety
- (01:25:02) Pentagon inks deals with Nvidia, Microsoft, and AWS to deploy AI on classified networks | TechCrunch
- (01:27:27) Google, Microsoft, and xAI will allow the US government to review their new AI models | The Verge
- (01:32:11) NSA Testing Anthropic’s Mythos to Find Flaws in Microsoft Tech
- (01:35:42) Introspection Adapters: Training LLMs to Report Their Learned Behaviors
- Research & Advancements
- (01:41:18) Recursive Multi-Agent Systems
- (01:51:47) Frontier Coding Agents Can Now Implement an AlphaZero Self-Play Machine Learning Pipeline For Connect Four That Performs Comparably to an External Solver
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.
Fler avsnitt av Last Week in AI
Visa alla avsnitt av Last Week in AILast Week in AI med Skynet Today finns tillgänglig på flera plattformar. Informationen på denna sida kommer från offentliga podd-flöden.