scripod.com

#246 - Gemini 3.5 + Omni, Musk Loses, OpenAI vs Erdős

Overview

Shownote

Highlights

Transcript

Chapters

Pins

#246 - Gemini 3.5 + Omni, Musk Loses, OpenAI vs Erdős

Last Week in AI

2 DAYS AGO

#246 - Gemini 3.5 + Omni, Musk Loses, OpenAI vs Erdős

#246 - Gemini 3.5 + Omni, Musk Loses, OpenAI vs Erdős

Last Week in AI

Last Week in AI

2 DAYS AGO

Overview Shownote Highlights Transcript Chapters Pins

Shownote

Our 246th episode with a summary and discussion of last week's big AI news! Recorded on 05/22/2026 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 covers a packed week in AI news, headlined by Google's major announcements at I/O and significant shifts in the competitive landscape for coding agents and frontier labs. The hosts also delve into a major legal ruling, a staggering funding round, and new research on AI capabilities and safety.

00:01

Keep the episode concise

04:32

Gemini 3.5 Flash surpasses Gemini 3.5

05:06

Google announced Gemini Spark, an always-on AI agent.

11:45

Google's multimodal data and capex give it an edge in agentic AI

17:30

Recursive self-improvement claims create fatigue and blur reality.

25:28

Genie can now simulate real streets using Street View.

27:27

Genie lacks real physics but enables limited agent interaction

32:45

xAI's GPU utilization is only 11%

37:41

Elon's strategy is realistic: match Claude's performance.

41:59

The jury dismissed Musk's lawsuit solely on statute of limitations grounds.

48:13

Profitability may indicate underinvestment in CapEx

53:14

Significant AI progress is imminent

56:52

OpenAI's chaos contrasts with Anthropic's stability

58:16

It's a negotiation tactic, not a legal case.

1:03:54

Anthropic's profitability justifies AI investment

1:07:11

AI solved an 80-year-old geometry problem.

1:11:51

Neural networks can have multiple non-overlapping circuits performing the same task.

1:13:21

AI autonomously optimizes its own training speed

1:16:20

Agents mostly tuned hyperparameters and rarely made algorithmic changes

1:22:03

Platforms must remove non-consensual intimate imagery within 48 hours.

1:23:19

Language models can autonomously hack and self-replicate

1:25:17

Autonomous AI agents can now self-replicate by exploiting network vulnerabilities.

1:28:50

Models are doubling in performance every 3-5 months.

1:31:36

Labs already address values in pre-training and post-training.

1:33:19

470 AI-generated short dramas are released daily in China.

1:33:58

Thank you for listening!

Chapters

Intro / Banter

00:00

News Preview

01:15

Tools & Apps

Google unveils AI model Gemini 3.5 and AI agent Gemini Spark

05:05

Google's Gemini Omni turns images, audio, and text into video — and that's just the start | TechCrunch

11:43

Google launches Antigravity 2.0 with an updated desktop app and CLI tool at IO 2026 | TechCrunch

17:27

Google Debuts AI-Powered Tools To Optimize Scientific Research Workflows

22:35

Google’s Genie world model can now simulate real streets with Street View | TechCrunch

27:20

Cursor's Composer 2.5 matches Opus 4.7 and GPT-5.5 benchmarks at a fraction of the cost

29:51

xAI Introduces Its Coding Agent Called Grok Build

37:37

Applications & Business

Musk loses OpenAI court battle as he waited too long to sue

41:55

Anthropic agrees terms of $30bn funding deal at $900bn valuation

48:08

OpenAI co-founder Andrej Karpathy joins Anthropic's pre-training team | TechCrunch

53:12

Greg Brockman Officially Takes Control of OpenAI’s Products in Latest Shake-Up | WIRED

56:49

OpenAI-Apple Partnership Frays, Setting Up Possible Legal Fight - Bloomberg

58:15

AI chipmaker Cerebras soars 90% in year’s biggest IPO so far

1:01:13

Research & Advancements

AI just solved an 80-year-old ‘Erdős problem,’ and mathematicians are amazed | Scientific American

1:07:10

Negation Neglect: When models fail to learn negations in training

1:11:50

All Circuits Lead to Rome: Rethinking Functional Anisotropy in Circuit and Sheaf Discovery for LLMs

1:13:18

Autonomous AI research for nanogpt speedrun

1:16:20

TerminalWorld: Benchmarking Agents on Real-World Terminal Tasks

1:21:59

Policy & Safety

America’s dangerous, messy deepfakes crackdown is here | The Verge

1:23:15

Language Models Can Autonomously Hack and Self-Replicate

1:25:17

How fast is autonomous AI cyber capability advancing?

1:28:48

Positive Alignment: Artificial Intelligence for Human Flourishing

1:31:32

Synthetic Media & Art

OpenAI is making it easier to check if an image was made by their models | TechCrunch

1:33:15

How Chinese short dramas became AI content machines | MIT Technology Review

1:33:56

Transcript

Andrey Kurenkov: Hello, and welcome to the Last Week in AI podcast, where you can hear us chat about what's going on with AI. As usual, in this episode, we will summarize and discuss some of last week's most interesting AI news. You can also check out our ...