scripod.com

#244 - GPT-5.5 Instant, Grok 4.3, OpenAI vs Musk

Overview

Shownote

Highlights

Transcript

Chapters

Pins

#244 - GPT-5.5 Instant, Grok 4.3, OpenAI vs Musk

Last Week in AI

3 DAYS AGO

#244 - GPT-5.5 Instant, Grok 4.3, OpenAI vs Musk

#244 - GPT-5.5 Instant, Grok 4.3, OpenAI vs Musk

Last Week in AI

Last Week in AI

3 DAYS AGO

Overview Shownote Highlights Transcript Chapters Pins

Shownote

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 andreyvkurenkov@gmail.com and/or hello@gladstone.ai Rea...

Highlights

This episode delivers a comprehensive, critical analysis of the most consequential AI developments from the past week—spanning model releases, safety incidents, industry shifts, policy evolution, and foundational research advances.

01:15

Today's AI agents operate in silos

04:40

Mythos led to Firefox fixing a large number of bugs

13:18

GPT-5.5 Instant outperforms GPT-5.4 on cyber-related benchmarks despite being lighter-weight

17:58

The 'goblin' obsession originated from over-rewarding creature metaphors in a nerdy model variant during RL training

27:00

Grok 4.3 sits on the Pareto frontier of cost per unit intelligence

30:51

XAI's Grok 5 is planned for later this year and targets 10 trillion parameters

39:27

Multi-agent orchestration enables a lead agent to delegate tasks to sub-agents

40:43

When demand is low, spare AI infrastructure capacity is used for 'dreaming' to keep GPUs active and optimize the compute fleet.

51:00

Greg Brockman's diary showed a focus on reaching $1 billion net worth, which contradicts his court statement of putting mission first

55:45

SpaceX provides compute to Anthropic, potentially transforming xAI into an AI cloud provider

58:48

Anthropic shows interest in SpaceX's orbital data center plan for its IPO

59:42

Anthropic is in talks to raise funds at a $900 billion valuation, a big jump from February

1:03:16

Asset managers can force AI tool usage in portfolio companies due to PE ownership

1:04:04

Forward-deployed engineers act as an accelerant to AI adoption, differing from self-serve API models

1:08:45

AI firms accounted for over a third of private-graded deals in 2025, up from 17% a few years ago

1:11:17

Risk doesn't disappear but migrates to private credit funds, insurers, and asset-backed markets

1:15:48

Anthropic's Natural Language Autoencoders map LLM activations directly to plain English explanations

1:19:23

Natural Language Autoencoders encode activation into plain English and decode it back, trained via reinforcement learning to minimize reconstruction error

1:22:06

The technology is already in production across OpenAI's superclusters.

1:27:26

The US government is considering pre-deployment licensing for AI models like Mythos

1:31:55

NSA is examining Anthropic's Mythos to find zero-day vulnerabilities

1:35:33

680 behavior adapters are trained for different proclivities, then introspection LoRa adapters teach the model to report those behaviors

1:40:59

Passing activations instead of text preserves chain-of-thought integrity in multi-agent systems

1:48:47

Frontier coding agents can implement an AlphaZero self-play pipeline for Connect Four under non-ML expert human direction

Chapters

Intro / Banter

00:00

News Preview

01:14

Response to listener comments

01:39

Tools & Apps

OpenAI releases GPT-5.5 Instant, a new default model for ChatGPT | TechCrunch

10:40

ChatGPT Became So Obsessed With Goblins That OpenAI Had to Intervene

15:23

xAI launches Grok 4.3 at an aggressively low price and a new, fast, powerful voice cloning suite | VentureBeat

24:14

Mistral's new flagship Medium 3.5 folds chat, reasoning, and code into one model

30:49

Anthropic updates Claude Managed Agents with three new features - 9to5Mac

36:28

ElevenLabs Revamps AI Music Platform as Fan-Focused Service

40:42

Applications & Business

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

41:57

Anthropic, SpaceX Sign Deal to Boost AI Computing Power for Claude Software - Bloomberg

52:28

Anthropic in talks with investors to raise funds at $900b valuation, higher than OpenAI

58:48

Anthropic and OpenAI are both launching joint ventures for enterprise AI services | TechCrunch

59:37

Anthropic and FIS Are Building an AI Agent to Help Banks Police Financial Crimes

1:03:15

AMD’s revenue jumps 38 percent from last year as Q1 data center sales hit $5.8 billion. | The Verge

1:04:02

Banks seek to offload risk to avoid ‘choking’ on data centre debt

1:05:51

DeepSeek could be valued at up to $50 billion in first fundraising, sources say | Reuters

1:11:08

Projects & Open Source

Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations

1:13:14

OpenAI just open-sourced its data center networking technology

1:19:23

Policy & Safety

Pentagon inks deals with Nvidia, Microsoft, and AWS to deploy AI on classified networks | TechCrunch

1:22:02

Google, Microsoft, and xAI will allow the US government to review their new AI models | The Verge

1:24:27

NSA Testing Anthropic’s Mythos to Find Flaws in Microsoft Tech

1:29:11

Introspection Adapters: Training LLMs to Report Their Learned Behaviors

1:32:42

Research & Advancements

Recursive Multi-Agent Systems

1:38:18

Frontier Coding Agents Can Now Implement an AlphaZero Self-Play Machine Learning Pipeline For Connect Four That Performs Comparably to an External Solver

1:48:47

Transcript

Andrey Kurenkov: This episode is brought to you by Progressive Insurance. Do you ever think about switching insurance companies to see if you could save some cash? Progressive makes it easy to see if you could save when you bundle your home and auto polici...