Verbalized Sampling: Prompting for Diverse LLM Outputs

Our 223st episode with a summary and discussion of last week's big AI news!

Recorded on 10/17/2025

Hosted by Andrey Kurenkov and co-hosted by Erik Schnultz

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:

• Anthropic and OpenAI have announced updates to their AI models and tools, including Haiku 4.5 and various business collaborations.
• Multiple companies like Slack and Salesforce are integrating AI assistants and agents into their platforms, enhancing task management and business operations.
• Recent research in reinforcement learning and agent memory curation highlights new methods for improving AI model performance and context management.
• California has passed a law to regulate AI chatbots for children and vulnerable users, and there are rising concerns over the increasing amount of AI-generated content on the internet.

Timestamps:

• (00:00:10) Intro / Banter
• (00:01:31) News Preview
• Tools & Apps
• (00:02:18) Anthropic launches new version of scaled-down ‘Haiku’ model
• (00:04:52) Everything OpenAI announced at DevDay 2025: Agent Kit, Apps SDK, ChatGPT, and more | ZDNET
• (00:09:11) Anthropic turns to ‘skills’ to make Claude more useful at work | The Verge
• (00:13:20) Microsoft launches ‘vibe working’ in Excel and Word | The Verge
• (00:17:22) Google releases Veo 3.1, adds it to Flow video editor | TechCrunch
• (00:19:40) Slack is turning Slackbot into an AI assistant | The Verge
• (00:22:52) Salesforce announces Agentforce 360 as enterprise AI competition heats up | TechCrunch
• Applications & Business
• (00:24:58) Broadcom stock pops 9% on OpenAI custom chip deal, adding to Nvidia and AMD agreements
• (00:27:58) How ByteDance Made China’s Most Popular AI Chatbot | WIRED
• (00:30:08) Amazon's Zoox Robotaxis Have Arrived In Las Vegas - Here's What Riders Are Experiencing
• (00:32:43) Waymo’s robotaxis are coming to London | The Verge
• (00:34:14) Reflection AI raises $2B to be America's open frontier AI lab, challenging DeepSeek | TechCrunch
• (00:35:58) General Intuition lands $134M seed to teach agents spatial reasoning using video game clips | TechCrunch
• (00:38:36) Supabase nabs $5B valuation, four months after hitting $2B | TechCrunch
• Projects & Open Source
• (00:40:58) Neuphonic Open-Sources NeuTTS Air: A 748M-Parameter On-Device Speech Language Model with Instant Voice Cloning - MarkTechPost
• (00:43:06) Anthropic AI Releases Petri: An Open-Source Framework for Automated Auditing by Using AI Agents to Test the Behaviors of Target Models on Diverse Scenarios - MarkTechPost
• Research & Advancements
• (00:44:25) [2510.13786] The Art of Scaling Reinforcement Learning Compute for LLMs
• (00:48:51) [2510.01171] Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity
• (00:51:22) [2510.12635] Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Tasks
• (00:54:31) [2510.07364] Base Models Know How to Reason, Thinking Models Learn When
• (00:57:24) [2510.12402] Cautious Weight Decay
• Policy & Safety
• (01:02:03) California becomes first state to regulate AI companion chatbots | TechCrunch
• (01:04:13) Over 50 Percent of the Internet Is Now AI Slop, New Data Finds
• Synthetic Media & Art
• (01:06:31) OpenAI Reverses Stance on Use of Copyright Works in Sora - WSJ
• (01:08:29) Character.AI removes Disney characters from platform after studio issues warning

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.