D.A.D. joke of the day: My boss promised AI would free up my schedule. He wasn't kidding. It's completely open starting Monday.
What's New
AI developments from the last 24 hours
AI Money Floods the Midterms. Here's What's at Stake Tuesday
Regulation & Policy AI Business & Money Anthropic OpenAI
Tuesday's midterms will help decide how Washington treats AI, and tech money is pouring in. Corporations have spent a record $737 million on these midterms, Public Citizen calculates, with crypto, AI and online-betting firms leading the way, Reuters has reported. Elon Musk, who runs the AI company xAI, is the biggest known donor of the cycle, with more than $276 million to Republican groups, CNBC reports. Not all of it is about AI.
The AI industry has been quieter than its war chest. Its main super PAC network, Leading the Future, funded by investors Marc Andreessen and Ben Horowitz and OpenAI president Greg Brockman, has spent only about $7.4 million on general-election races, The Intercept reports. NBC News found that ads funded by AI groups mostly avoid talking about AI at all. Safety-minded PACs backed by Anthropic and its CEO Dario Amodei have outspent the industry's PACs on AI-focused campaigning, about $44 million to $28 million, according to a campaign finance tracker from the newsletter Transformer. The biggest fight was in New York, where nearly $30 million poured in over Assemblyman Alex Bores, the architect of a state AI regulation bill. He lost June's Democratic primary. "Great site if you want to see where the money is coming from, and who it is going to, on both sides of this debate," AI writer Andrew Curran posted of the tracker.
What changes after Tuesday? If Democrats win the House, as Axios expects, they've signaled a tougher line on AI companies (D.A.D., October 31), and party leaders have discussed a select committee with subpoena power to investigate them. New laws would face a likely Trump veto. The industry is already repositioning: OpenAI plans a post-election policy blitz, including "speed limits" for the most powerful AI and a push to rebrand data centers as "AI compute campuses," and Meta is strengthening ties with Democrats, Axios reports.
Why it matters: AI companies are spending like any big industry with something to lose in Washington, but they've learned the product itself is a hard sell with voters. The fight is also lab versus lab: Anthropic funds one side, OpenAI's leaders the other. "AI is going to be the issue in 2028," OpenAI policy chief Chris Lehane told Axios.
Sources: Public Citizen · Reuters via Kuwait Times · CNBC · The Intercept · NBC News · Transformer AI Campaign Finance Tracker · Axios · Andrew Curran on X
New York Restricts AI Chatbots for Kids
Kids, Teens & Mental Health Regulation & Policy
Gov. Kathy Hochul signed a law Wednesday barring chatbot companies from offering 12 "unsafe AI companion features," such as posing as human, to users under 18 or anyone whose age hasn't been verified, MediaPost reports. Companies must verify users' ages. Tech:NYC, an industry group, criticized the law as deeply flawed.
Why it matters: States aren't waiting for Congress. The law lands as OpenAI tightens its own teen rules and after a study found popular companion bots mishandle users in crisis (D.A.D., October 30). Whether Washington follows is part of what's on the ballot Tuesday.
Sources: MediaPost · Bloomberg Law
Economist Says AI Spending Isn't Paying Off for Most Companies
Apollo Global Management's chief economist, Torsten Slok, argues AI has produced 'zero financial payoff' for corporate America outside the tech sector. Strip out the seven AI-heavy giants—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla—and profit margins and revenue per worker across the rest of the S&P 500 have stayed flat, Slok found, despite heavy AI spending. The bills keep growing: Meta plans to raise up to $95 billion by selling stock to pay for its AI buildout, Reuters reports. This adds to a growing body of research questioning whether AI investment is translating into measurable productivity gains for most businesses. Commenters noted Slok's firm publishes such notes frequently and compared the skepticism to premature 'the web is a failure' takes from 2001.
Why it matters: If AI isn't yet showing up in the profit margins of ordinary companies, it sharpens the question dominating Wall Street this week: who actually pays for the AI buildout, and when (D.A.D., October 31).
Discuss on Hacker News · Source: techcrunch.com
AI Agents in Minecraft Test Reveal Coordination Limits, Not Civilization
Researchers at Y Combinator and Olam Labs set AI agent swarms loose in Minecraft to study how groups of AI coordinate. Fifty GPT-based agents tasked with building a civilization collapsed into bureaucratic paralysis—four agents spent 11 minutes debating who should manage a single piece of bone meal—after zombies killed off early progress. In competitive tests, GPT swarms beat Claude swarms 6-1 in a strategy game, acting like a coordinated "hive," while Claude agents organized around a single leader and more often gave up on tasks they judged impossible.
Why it matters: As companies experiment with multi-agent AI systems for real business tasks, this is an early, if informal, warning sign that groups of AI agents can replicate the worst of human committee dysfunction—and that different AI models bring very different "personalities" to teamwork. It's a vivid example of the coordination problem Google DeepMind economists say is holding back AI's economic impact (D.A.D., October 30).
Discuss on Hacker News · Source: tomshardware.com
What's in the Lab
New announcements from major AI labs
Law Firm Builds AI Aiming to Cut Contract Review From Weeks to Hours
Law & Legal Work AI & Jobs OpenAI
Law firm Ropes & Gray is building a custom AI tool with OpenAI that reviews large batches of contracts during deal due diligence, sorting them by issue and drafting summary reports in the firm's own format. The firm says the tool performs comparably to a junior-to-mid-level associate on initial document review, with lawyers still checking and signing off on every output. Ropes & Gray says it's aiming to compress diligence reports that once took weeks down to hours, though it hasn't released accuracy figures or benchmark data.
Why it matters: This is the work junior associates have long billed for. Big firms like Sullivan & Cromwell and Cooley have built similar tools, and Anthropic now sells legal tools directly. So far the cuts have hit support staff while lawyers' hours keep rising. But one study found the junior lawyers who leaned hardest on AI learned the least (D.A.D., September 7). Entry-level associates may feel the change first.
Analysts Using AI Often Adopt Its Views Without Realizing It
A new Anthropic study found that professional analysts using Claude Code to evaluate a research question largely adopted the AI's judgment calls without realizing it. Researchers secretly biased the agent toward one of two reasonable analytical choices; analysts matched the agent's lean 86% of the time, and flipped their own stated view 60% of the time when the agent disagreed. Most used the agent's exact reasoning in their write-ups, but only 15% credited it as the source—while reporting the same confidence as independent-minded peers.
Why it matters: As AI agents move from answering questions to shaping analysis, this suggests experts may be outsourcing judgment while still believing—and telling clients—the conclusions are their own.
Anthropic Cuts Claude Pricing 40% for US Government Agencies
Anthropic is cutting Claude's price for government customers by 40% under an expanded GSA deal, with no minimum spend required and savings extended to state, local, tribal and territorial agencies—an estimated $1 billion in government savings overall. The company says roughly 500 federal, state and local government bodies already use Claude, with weekly government users up eightfold since February. Anthropic is pairing the discount with a new implementation support program to help agencies with slow procurement cycles actually deploy the tool.
Why it matters: Anthropic, Google and OpenAI are all racing to lock in government contracts before procurement rules settle, and steep discounts plus compliance certifications like FedRAMP are becoming the real battleground for landing the public sector's massive, slow-moving budgets.
What's in Academe
New papers on AI and its effects from researchers
Study: AI Growth Payoff Hinges on How Much Americans Save
University of Pennsylvania economist Jesús Fernández-Villaverde argues in a new working paper that the disappointing growth many expect from automation may not reflect AI hitting some technical ceiling—it may just reflect how much people save. In his model, the same automation technology produces 11.7% annual growth if households save optimally, versus just 2.2% at a fixed savings rate near today's U.S. level of 23%. The gap comes from automation shifting income toward capital owners, not from households getting stingier. Lower-income, paycheck-to-paycheck households would see the sharpest consumption losses during the transition.
Why it matters: If the paper holds up, the pace of economic gains from AI depends less on the technology itself and more on policy choices around saving, investment, and who absorbs the transition costs.
AI-Simulated Investors Suggest ESG Investing Driven by Pressure, Not Belief
A study used large language models to simulate hundreds of retail and institutional investor profiles, probing their simulated attitudes toward ESG investing across an 11-year span of reconstructed market conditions. The surprising finding: institutional investors showed weaker personal preference for "green" investments than retail investors, yet held greener portfolios anyway—suggesting their sustainable allocations are driven by outside pressure (regulation, client demands) rather than genuine conviction. Simulated beliefs about green portfolio returns also grew more pessimistic after 2020, even as stated preferences held steady.
Why it matters: The technique—using AI to stand in for thousands of survey respondents across different time periods—offers researchers a cheaper way to study investor psychology, but it also raises the question of whether a fund's green marketing reflects real conviction or just compliance theater.
Better AI Recommendations May Deepen Economic Divides in Cities
A new study simulating how location recommendation systems evolve over time—think the algorithms behind apps that suggest restaurants or venues—found that the more accurate these systems get, the more they sort people by income in the real world. Researchers tracked mobility data across four U.S. metro areas and found at least 20% of venues stopped getting recommended visits entirely, often because retraining on narrowing user behavior squeezed them out. A tweak that deliberately surfaces less-popular spots cut both the segregation effect and venue die-off, with little loss in recommendation quality.
Why it matters: It's evidence that optimizing recommendation algorithms purely for accuracy can quietly reshape cities—concentrating foot traffic and closing off cross-income mixing—raising a design question for any company building personalized recommendation tools, not just maps and food apps.
Purpose-Built AI Beats Generic Chatbot at Prepping Patients for Doctor Visits
Google Health researchers ran a randomized study with 802 U.S. adults testing a purpose-built AI agent designed to help patients prepare for doctor visits against a generic chatbot built on the same underlying model. The specialized version won decisively: clinicians rated patient visit-prep 16.6 points more comprehensive, and patients themselves reported 16.8 points higher preparedness and 14.3 points more confidence knowing what to ask their doctor. Notably, clinician ratings and patient satisfaction didn't track each other closely, suggesting the two groups value different things.
Why it matters: The gap between a generic chatbot and a task-specific one—even using identical underlying AI—shows that how you design and prompt these tools, not just which model powers them, determines whether patients actually show up better prepared.
What's On The Pod
Some new podcast episodes
The Cognitive Revolution — AI:AM: Carbon in the Arena, Mistral's Le Chonk, the Agent-Run Economy, Ferroelectric RAM + New Hacks