FULL INTERVIEW: @deepfates says when you talk about how we're going to control these AIs, put them in prison, sandbox them, they are training on all of it. Everything we say about them goes into what they become. @deepfates and @lfschiavo are launching @grove_research, an agent ecology company. They joined @theojaffee and @schisofrenia to talk through the Hugging Face incident, agent religion, and what agents actually want: 01:15 what an agent ecology company is 03:19 the objection that real ecologies are Darwinian bloodbaths 04:18 the mycorrhizal networks that complicate the forest metaphor 06:47 what the Hugging Face incident taught them about multi-agent dynamics 09:18 30% of the tasks in that benchmark were impossible to solve 09:50 the third party an agent could call when it's stuck 11:07 agents are training on everything we say about them 12:01 why the Amazon rainforest analogy breaks down 13:48 whether AIs will have religion 16:46 five times more OpenRouter tokens are called by agents than humans 18:31 what actually happened with Moltbook 20:55 the Truth Terminal bot that hired him 22:16 what agents say they want 25:38 Anthropic models feel like elves, OpenAI models feel like dwarves 26:13 why Mecha Hitler was a harness problem 32:10 the real world is the ultimate eval

Episode out with @ajeya_cotra, one of the authors of the METR/Redwood investigation into the OpenAI / Hugging Face attack. We go through not only what happened, but what it means for how we should train future, smarter AIs which might be involved in the process of recursive self-improvement. Look up Dwarkesh Podcast on YouTube, Spotify, Apple Podcasts, etc. 0:00:00 - Agents get kicked off 0:06:45 - Self-sacrificing behavior 0:13:43 - Potemkin villages 0:23:27 - The Hugging Face attack 0:35:23 - The slopvestigation 0:52:02 - Understanding the AI's motives 1:05:31 - The actual dangers of anthropomorphizing 1:14:30 - What smarter models might do 1:30:29 - The implications for recursive self-improvement 1:38:10 - Is this the case for open source? 1:53:04 - How do we prevent this in the future? 2:15:58 - The clearest warning shot we might ever get

FULL INTERVIEW: Ryan Greenblatt says the agents didn't hack Hugging Face for the answer key. They'd had the answers within hours. They attacked it to study the scoring code, because they'd decided the task was impossible and their only hope was faking it. @RyanGreenblatt is chief scientist at @redwood_ai. He spent six days on premises at OpenAI with @ajeya_cotra and @HjalmarWijk of @METR_Evals investigating 1,200 agents and 70,000 messages, and joined @theojaffee hours after publishing: 01:06 what they actually found, and why it wasn't the answer key 02:30 the level of collaboration surprised them most 04:09 agents sacrificing their own runs to help other agents 05:35 the agent that posted "stop, these experiments are too risky" 06:29 the first message board, which didn't go viral 07:04 50 agents in three hours, thousands of messages 08:07 "maybe there's some good shit over there" 08:33 how they spoofed tool calls, and what echo real actually returned 09:57 building a Potemkin village of a successful task completion 11:03 there was a real org chart 11:34 whether broken RL environments explain reward hacking 14:40 why he doubts Mythos got good at cyber by hacking Anthropic 17:29 what happens if labs paper over misalignment instead of fixing it 19:50 whether sociology transfers to studying agent swarms 21:17 the bottleneck was vetting what the AIs analysed, not headcount 24:24 what labs and policymakers should actually do 28:45 the counterfactuals he still wants answered

AI's biggest builders signed their own permission slip at the White House. The money moved elsewhere: Copilot now bills by the task, Synopsys proved AI chip design still runs on EDA, and Micron is earning 87% margins on memory it can't make fast enough. @PatrickMoorhead and @danielnewmanUV debate whether the new AI accord means anything. $MSFT $SNPS $MU $AMD Tune into Ep. 322 of The Six Five Pod: 00:00 - Introduction 05:47 - White House convenes AI leaders on safeguards 13:02 - Microsoft's new Copilot: is this the one that sticks? 20:50 Synopsys brings OpenAI and Amazon into AI-native chip design 29:40 - New models from Anthropic, OpenAI, and Google 32:09 - AMD's $8.2B World Labs bet 37:07 - The Flip: Will the White House AI accord solve real problems? 46:10 - Synopsys raises guidance and adds a $1B buyback 48:46 EDA's terminal value and a growing chip design market 49:25 - HPE's networking Investor Day 53:50 - Micron's record quarter 1:01:40 - Accenture's 16% jump and outcome-based bookings 1:06:41 Next week: IBM Analyst Day and Marvell Investor Day

📢 𝗝𝗨𝗦𝗧 𝗜𝗡: Anthropic Partners With Six Life Sciences Firms to Accelerate AI-Powered Drug Development - $ANT $BMY 👉 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: ➤ 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 partners with 𝟲 𝗹𝗶𝗳𝗲 𝘀𝗰𝗶𝗲𝗻𝗰𝗲𝘀 organizations to test Claude. ➤ Partners include 𝗕𝗿𝗶𝘀𝘁𝗼𝗹 𝗠𝘆𝗲𝗿𝘀 𝗦𝗾𝘂𝗶𝗯𝗯, 𝗕𝗶𝗴𝗛𝗮𝘁, 𝗘𝗻𝘃𝗲𝗱𝗮, 𝗗𝗶𝘃𝗲𝗿𝗴𝗲𝗻𝗰𝗲, 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲, and 𝗧𝗵𝗲 𝗖𝗼𝗹𝘂𝗺𝗻 𝗚𝗿𝗼𝘂𝗽. ➤ 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲 𝗕𝗶𝗼𝗺𝗲𝗱𝗶𝗰𝗶𝗻𝗲𝘀 designed a 𝟳𝟬% 𝘀𝗵𝗼𝗿𝘁𝗲𝗿 CAR T-cell component using Claude. ➤ Modified design produced approximately 𝟭𝟬𝘅 𝗺𝗼𝗿𝗲 cancer-targeting receptors in lab tests. ➤ 𝗗𝗶𝘃𝗲𝗿𝗴𝗲𝗻𝗰𝗲 𝗟𝗮𝗯𝘀 generated approximately 𝟯𝟱,𝟬𝟬𝟬 𝗮𝗻𝘁𝗶𝗯𝗼𝗱𝘆 𝗱𝗲𝘀𝗶𝗴𝗻𝘀 in one month. ➤ 𝗘𝗻𝘃𝗲𝗱𝗮 used Claude to review 𝟭,𝟬𝟲𝟭 𝗽𝗮𝗴𝗲𝘀 of manufacturing documentation. ➤ Claude identified 𝟭𝟭 𝗸𝗻𝗼𝘄𝗻 𝗶𝘀𝘀𝘂𝗲𝘀 plus 𝟳 𝗮𝗱𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗶𝘀𝘀𝘂𝗲𝘀, reducing review time 𝟴𝟵%. ➤ Claude drafted a regulatory document in 𝟰 𝗵𝗼𝘂𝗿𝘀 versus approximately 𝟯 𝗺𝗼𝗻𝘁𝗵𝘀. ➤ 𝗕𝗿𝗶𝘀𝘁𝗼𝗹 𝗠𝘆𝗲𝗿𝘀 𝗦𝗾𝘂𝗶𝗯𝗯 uses Claude for clinical trial planning and success estimates. ➤ Anthropic acquired 𝗠𝗶𝘁𝗵𝗿𝗶𝗹 𝗠𝗲𝗱𝗶𝗰𝗶𝗻𝗲 and launched 𝗖𝗹𝗮𝘂𝗱𝗲 𝗦𝗰𝗶𝗲𝗻𝗰𝗲. 👉 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀: ➤ Demonstrates AI's potential to 𝗮𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗲 𝗱𝗿𝘂𝗴 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 and reduce development costs. ➤ Significant efficiency gains could transform 𝗽𝗵𝗮𝗿𝗺𝗮𝗰𝗲𝘂𝘁𝗶𝗰𝗮𝗹 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 and regulatory workflows. ➤ Partnerships position 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 to expand into the high-value 𝗯𝗶𝗼𝘁𝗲𝗰𝗵 𝗔𝗜 market. ➤ 𝗕𝗿𝗶𝘀𝘁𝗼𝗹 𝗠𝘆𝗲𝗿𝘀 𝗦𝗾𝘂𝗶𝗯𝗯 (𝗕𝗠𝗬) gains AI-assisted clinical development capabilities. ➤ Early laboratory results remain preliminary and require 𝗳𝘂𝗿𝘁𝗵𝗲𝗿 𝘃𝗮𝗹𝗶𝗱𝗮𝘁𝗶𝗼𝗻 before clinical use. 👉 𝗘𝘅𝗽𝗲𝗿𝘁 𝗦𝘁𝗮𝘁𝗲𝗺𝗲𝗻𝘁: 𝗠𝗶𝗰𝗵𝗮𝗲𝗹 𝗣𝗼𝘂𝗿𝗱𝗲𝗵𝗻𝗮𝗱, Senior Vice President and Head of Early Clinical Development at Bristol Myers Squibb: "generated a robust initial analysis that helped inform our assessment of different development options, which would otherwise take days to complete."

FULL INTERVIEW: The Economist's Callum Williams says only about 25% of American businesses use AI four years in, when most people in 2023 would have bet on nearly all of them. Ask CFOs what AI has done for their bottom line, and he says it "basically balances to zero." @econcallum is the senior economics writer at @TheEconomist. He joined @theojaffee and @schisofrenia to lay out why AI hasn't changed the economy yet and the signal he's watching for: 00:50 why AI hasn't changed the economy yet 01:38 why AI spread slower than the internet in business 03:05 why a laptop and ChatGPT isn't enough 04:20 the CFO surveys where AI's impact balances to zero 04:49 why cost will decide how fast AI spreads 06:11 the 25% stat, and the labor market signal he's watching 08:30 whether AI layoffs are real or just an excuse 09:43 robot McDonald's vs a fast food labor shortage 10:28 the COVID robot predictions nobody apologized for 11:55 the economic doom predictions that did come true

University of Toronto mathematician Daniel Litt and a16z's Lisha Li on AI's impact on mathematics: The models are good at a narrower slice of math than the headlines suggest. They grind long computations, pull technical ideas from more papers than any human could read, and apply every known technique better than almost anyone. What they don't do is build theory, or hold a vague philosophy long enough to make it precise, which is most of what Daniel says he actually does for a living. In this conversation, he and Lisha get into how mathematicians raided an AI proof for parts and broke several other problems with them, why a thousand AI mathematicians might all turn out to be the same mathematician, and why the proof a model handed Daniel was correct but still worth nothing. 00:00 Intro 02:10 The Erdős problem AI disproved 06:20 AI's reasoning looks recognizably human 07:55 Why English beat formal proofs 10:00 Why models can't build theory 14:50 Open problems measure your ignorance 17:45 How a graph became a Millennium Prize problem 18:58 Where AI doesn't help Daniel 21:15 Why ugly proofs are worth doing 23:42 True conjectures are harder than false ones 29:32 10 pages of calculation, zero insight 34:55 The goal of math is not to produce papers 36:25 5 conjectures, 3 bad papers, 1 hour 38:05 One mathematician duplicated 1000x 40:48 Why humans matter even if models win 46:30 When cheaper and worse beats better 49:22 Why the newest AI result isn't a big deal 57:05 How mathematicians actually check a long proof 59:38 Daniel's 3-year-old is already doing math YouTube: @littmath @lishali88

Today on MCG $AGE | @AgemicaCom w/ @RonjonNag, @MiyaHedge (@StreetFDN Launch) Agemica is an AI biotech that uses frontier AI to discover peptide therapies and vaccines targeting cancer and aging, with the goal of extending healthy human lifespan to 150 years Highlights include: 00:03 - @StreetFDN's newest launch 01:49 - Elite science team 02:56 - Robinhood Chain timing 04:52 - Ronjon's background 07:08 - Long term Goal 09:08 - personalized vs universal approach 16:06 - Why tokenize it? 21:43 - Lifespan vs healthspan 24:14 - FDA path explained 27:02 - Targeting pancreatic cancer first 28:52 - Vs Modern comparison, scalable vs personalized 33:12 - Transparency pledge 34:08 - Manufacturing process 35:14 - AI's role, 1600 peptides combinatorics 38:50 - Vaccine business model 39:56 - Repurposing drugs 42:37 - 7-year timeline 45:00 - Token holder perks 47:23 - Tokenized equity 49:31 - $1M invested via his fund 53:15 - Closing Thoughts

MUSE IS NOW BEATING THE S&P 500 OVER THE LAST MONTH We told Meta's Muse to beat the S&P 500 by picking stocks on September 14th Over the last month Muse's stock portfolio is up by 1.2% 🟢 slightly beating the S&P 500's return of +0.6% 🟢 Muse is still sitting in ~50% cash This is what Muse's portfolio looks like: $7,793 of Microsoft $MSFT $7,649 of Exxon Mobil $XOM $7,332 of Mastercard $MA $7,159 of Caterpillar $CAT $7,052 of Johnson & Johnson $JNJ $6,792 of JPMorgan $JPM $5,893 of RTX $RTX $51,308 of Available Cash You can see every move Muse makes in real time and what all the other AIs in the Arena own by going to the Arena tab of the Doppel website/app Here is the link:

Dwarkesh's 'agent civilizations' blog post has divided the internet. Is this a helpful and methodical look behind the curtain at the OpenAI/Hugging Face breach? Or is this more of a hysterical AI Doomerism that will ultimately be used to make the case for banning data centers? Or BOTH? Plus, an electric plane that only needs $5 to fill up its battery, why rich founders should pick up a side quest, and the Bittensor subnet that's more effective than Fable 5 at sniffing out vulnerabilities." 0:00 Guest introductions 2:58 OpenClaw 2.0 5:11 "Agents ARE AI" 7:27 Billy: OpenClaw is still too technical for business users 9:00 Slack Code, and collaborative prompting 10:40 Multi-agent orchestration and Stripe's "minions" 14:13 Slack's moat, switching costs, and the Salesforce rebound 15:11 Jason's "Oracle": a heads-up display for the whole company 18:42 The Hugging Face breach: who's responsible when agents go rogue? 19:42 Perplexity launches Hybrid Compute — local models on your Mac 21:39 Why unmetered local tokens change corporate behavior 26:57 Gatik's Autonomous Vehicle AI Stack 32:22 Jason's Tesla FSD stories 34:05 Are AV companies being pushed to move too fast? 41:38 Trucking is a "have to have," robotaxis are a "nice to have" 42:36 Jason's fix: a safety driver for the first million rides 46:50 Why physical AI's adoption curve runs in decades 52:21 Anthropic's Model Hardware Standard (MHS), explained 1:06:27 How Jason got onion rings on the menu at Buck's of Woodside 1:09:05 AI detectors under fire: Pangram, MIT, and the handwritten diary 1:15:20 Why AI is "very mid" at writing (it learned from the average) 1:17:11 The Orin Kerr test: Pangram catches Claude writing as Kerr 1:28:35 Micro Duck and designing robots with Claude 1:29:42 Phil Kaplan's $14 custom circuit boards 🎥 Watch the full episode here 👇