Our 245th episode with a summary and discussion of last week's big AI news!

Recorded on 05/13/2026

Feel free to email us your questions and feedback at andreyvkurenkov@gmail.com and/or hello@gladstone.ai

Read out our text newsletter and comment on the podcast at https://lastweekin.ai/

In this episode:

OpenAI released new voice intelligence API features including GPT Realtime 2 (GPT-5-powered) plus realtime translation and Whisper transcription, emphasizing the latency–reasoning tradeoff, larger context, and new guardrails amid fraud risks.
Thinking Machines previewed a low-latency, full‑duplex conversational system with a two-model architecture and custom inference stack, reporting strong interactivity benchmark results but without public access or third‑party validation yet.
Anthropic pushed further into vertical products with Claude for Legal and deeper AWS availability, while ongoing ecosystem tension grows as platform model providers compete with application-layer companies.
Safety, policy, and research updates included OpenAI’s self-harm trusted contact feature, Anthropic work on reducing agent misalignment by training ethical “why” reasoning, OpenAI’s investigation of accidental chain-of-thought grading in RL, and Meta horizon eval updates showing benchmarking limits for long task horizons.

Timestamps:

Tools & Apps

Applications & Business

(00:47:38) Sam Altman was winning on the stand, but it might not be enough | The Verge
(00:55:04) Nvidia C.E.O. Jensen Huang Hitches Ride With Trump to China After Last-Minute Invite - The New York Times
(00:58:40) AWS expands Anthropic partnership with Claude Platform launch
(01:01:13) Chinese grey market sells Claude API access at 90% off by using stolen credentials, model substitution, and harvesting users' prompts and outputs for resale as AI training data — 'transfer stations' operate through proxy networks that harvest user data
(01:06:43) DeepMind Spinout Isomorphic Labs Raises $2.1 Billion to Design Drugs With AI - Bloomberg

Projects & Open Source

Policy & Safety

(01:14:04) Teaching Claude why
(01:21:45) Import AI 455: Automating AI Research
(01:28:31) ChatGPT's New Safety Feature Could Alert 'Trusted Contact' to Risk of Self-Harm - CNET
(01:30:09) Investigating the consequences of accidentally grading CoT during RL
(01:34:46) Natural Language Autoencoders criticism
(01:39:15) Review of the "Risks from automated R&D" section in the Anthropic Risk Report (February 2026)

Synthetic Media & Art

Research & Advancements

#245 - TML-Interaction, Claude For Legal, Sam Altman on Stand

Last Week in AI