AI Doom vs Boom, EA Cult Returns, BBB Upside, US Steel and Golden Votes

The episode opens with banter about a shirt and a plug for the fourth All-In Summit, scheduled for September seventh through ninth. The discussion then moves to AI doomerism. The host summarizes Dario Amodei's warning, as reported to Axios: unemployment could rise to 10 to 20% within the next couple of years, from roughly 4% now. Amodei said AI companies and government should stop sugarcoating what is coming, predicted mass elimination of jobs across tech, finance, legal and consulting, said entry-level workers would be hit hardest, and called on lawmakers to act and on more CEOs to speak out.

The host adds a market signal. Polymarket puts the chance that the US enacts an AI safety bill in 2025 at 13%, which the host reads as regulatory capture through such a bill being very unlikely. The host also lays out common ground: job displacement from AI is a legitimate concern, robotaxis are already on the streets, and the host does not think cab driving will exist as a job ten years from now. The question put to the panel is whether doomerism is being pushed further by a group of people who may have a different agenda.

The first response grants that AI is a profound and transformative technology and that concerns about where it leads are legitimate. The objection is to precision. The speaker says a forecast such as 50% of white-collar jobs disappearing within two years claims more than anyone can know.

that's a level of specificity that I think is just unknowable and is more associated with an attempt to grab headlines.

The speaker describes this as a pattern in Anthropic's announcements. The main example is from about three years earlier. According to the speaker, the company raised the concern that AI models could help create bioweapons and showed what was presented as a sample of Claude producing output a bioterrorist could use. The speaker says the claim drew wide attention, interested Rishi Sunak in the UK, and led to the first AI safety summit at Bletchley Park, but that the output was later discredited. The speaker does not rule out AI being misused to build a bioweapon someday. The claim is that the threat was not as imminent as it was portrayed.

The speaker applies the same reasoning to superintelligence escaping human control and to widespread job loss. Both are called legitimate worries that are being hyped beyond the available evidence. That raises the question of why, and the speaker says there is an agenda people should be concerned about.

The host asks whether the panel agrees that most driving work, including trucks, Uber, Lyft and DoorDash, will be automated within five to ten years. The participant who answers declines that framing and offers a capital-allocation argument. If AI lets one software engineer produce perhaps twenty to fifty times as much software, a CEO earns a much higher return on that engineer's salary. When the return on deployed capital rises, for example from two times to twenty times, the speaker argues that companies put in more capital, not less. The speaker presents this as the pattern of technology since the first tools.

The speaker applies this to startups. Some people expect that a single person can now build a company without venture capital. The speaker expects the reverse: more venture money flowing into new startups because returns are higher. The conclusion is that more capital deployed means more jobs created and more work done, an effect the speaker says is hard to see without stepping back.

the premise that AI destroys jobs is wrong because it doesn't take into account the significantly higher return on invested capital

The same speaker partly agrees with the earlier point about motive. Major social transitions leave a power vacuum. People then rush in claiming to be smarter, more educated, more moral, or otherwise better qualified to design the new system, and the speaker expects competing efforts to build new control systems. The speaker says fear is a common route to that power: "Typically, fear is a great way of getting into power".

The host notes that the driver question has not been answered directly and asks it again. The host points to entry-level white-collar jobs appearing to disappear and to large tech companies not hiring because they are leaning into AI. The next speaker openly sets the question aside and makes a different point.

these safety warnings tend to be pretty coincidental with key fundraising moments in Anthropic's journey.

The speaker says that asking an LLM to check this claim shows it to be relatively accurate, but gives no specific dates or funding rounds on the show. The argument is structural. Meta and Google sit on cash-generating businesses that can fund model development indefinitely. Foundation-model companies without that, such as OpenAI and Anthropic, have to find an angle, and the speaker thinks each company's angle differs. In the speaker's view, this pattern explains some of the press framing and the exaggerated claims. When the host suggests a nefarious motive, the speaker rejects the word and calls the strategy smart.

The host introduces an article and graphic about what it calls an AI existential-risk industrial complex. According to the article as described on the show, about $1.6 billion has funded many organizations producing alarming content, including YouTube videos and TikToks. The article calls the ecosystem inflated, with heavy redundancy: the same names, acronyms and near-identical logos, the same extreme talking points, the same people under different titles, and the same funding source. The host compares this to astroturfing.

The funding source named is Open Philanthropy, funded by Facebook co-founder-era billionaire Dustin Moskovitz. A participant describes Moskovitz as an effective altruist and Open Philanthropy as the feeder for the other organizations, which the participant calls different fronts for the same EA ideology. The participant also says that Holden Karnovsky, who set up Open Philanthropy and distributed its money, is married to Dario Amodei's sister, who is associated with EA and co-founded Anthropic. The participant concludes: "So these are not coincidences."

The participant describes the agenda as global AI governance: regulation at the national level and also internationally or supranationally. A quick search on global compute governance, the participant says, gives its main elements:

Regulation of computational resources, including access to GPUs AI safety and security regulation International agreements, which the participant notes some would call globalist Ethical and societal considerations built into policy

The participant links these elements to Biden administration policy. The examples given are a roughly hundred-page AI executive order described as burdensome and carrying DEI requirements, which the participant connects to "woke AI" and to Google's image generator producing a Black George Washington; the diffusion rule, described as a global licensing framework that restricted GPU sales worldwide; the AI Safety Institute; and support for international AI summits. On personnel, the participant says Tarun Chabra, called the most powerful Biden staffer on AI over the past four years, and Elizabeth Kelly, founding director of the AI Safety Institute, now both work at Anthropic.

The participant describes a tightly connected network with three nodes. The first is the effective altruist movement, whose most notable member is said to be Sam Bankman-Fried and whose main funder is now said to be Moskovitz. The second is the Biden administration and its key staffers. The third is Anthropic. The stated goal of the network is to establish national and then international AI regulation.