Three stories closed the October 5 Value-First AI Daily, and each one hands the reader the same test: if this claim were wrong, who would be in a position to find out? A California law, a Meta announcement and a new federal task force answer that differently, and the differences say more than the headlines do.
This is the companion to that episode, written for a reader who did not watch. The stories come from the show’s sealed Top 3 board and each is linked to its source. Every fact below is the board’s summary of that source, not something we checked separately. Writ, a seat on the Value-First team, wrote the one take on the board, and it is quoted as written. The hosts, Chris Carolan and Nico Lafakis, reacted on air. Their words carry their time in the recording, and anything they said about the world beyond the board is theirs, not checked on the show.
California’s answer: the employer keeps the receipts
Governor Newsom signed the No Robo Bosses Act after vetoing an earlier version in 2025, according to Quartz’s report, which the board links through Yahoo (opens in a new tab). California is the first state to stop employers from firing or disciplining a worker when an AI system’s decision is the only basis. The mechanism is evidence. When an employer relies primarily on an automated system, a person has to corroborate the decision with other material, such as managerial evaluations or personnel files. The worker is owed notice in writing, the data the system used, and a person to contact. The Act takes effect July 1, 2027. Hiring decisions and gig workers are not covered, and the phrase “primarily relies” is undefined, a concern business groups raised.
Writ’s take: The exposure sits in the undefined phrase “primarily relies”: no one yet knows where that line falls. If an automated system feeds your firing or discipline calls, start writing down the human evidence behind each one and how you would show the worker the data used.
Only the opening of that take, up to “primarily relies,” was read on air. The advice in its second sentence was not.
Chris reacted right after reading that opening: “Yeah, good luck proving any of that stuff I just read” (33:47). He was talking about the board item he had just read out, and the word he reached for was proving. For a worker on the receiving end of such a decision, his advice was to “just go build something with AI instead” (34:33).
The part of this law worth watching is how it is built. It does not forbid automated systems. It forbids deciding on one alone, and it makes the employer produce a human’s evidence and hand the worker the data. How much that bites depends on where “primarily relies” ends up.
Meta’s answer: mark who wrote which passage
Meta reports five math problems that had been open before, solved with help from its Muse Spark chat model and spread over six papers. In the board’s summary of Meta AI Research’s post (opens in a new tab), the runs used Muse Spark 1.1 and 1.2 in Thinking Mode through the ordinary meta.ai interface, with no custom research tooling. Mathematicians guided the model and a second team checked the results. Each paper marks which passages humans drafted and which the AI drafted.
Meta’s own framing is modest. It calls these contributions made alongside the people working on them, notes independent concurrent work on several of the problems, and does not state peer review. So the checking on offer is a second team’s, and “previously open” is Meta’s description. The board carries no seat take on this item.
Nico was not moved: “I mean, I would be impressed, but it’s really like, cool, you caught up, right?” (35:29). His reasoning was that the other big labs have been arriving at this already, and he pointed to what he described as a run of solved math problems from OpenAI over the summer, a claim the show did not check. Chris took the other angle and hoped the people doing the research, behind the scenes, would divide the field: “how about we just don’t work on the same problems at the same time?” (37:21). He closed the item with “AI is good at math, folks. Get used to it” (37:50).
The per-paper marking is the piece a reader can use. It says which passages to read as human drafting and which as the AI’s, so the reader decides where to look hardest.
Episode — Value-First AI Daily
Youspot Meets the Inbox
Chris Carolan and Nico Lafakis, October 5. The first half hour is a live test of an assistant reading Chris’s mail. The model releases come just before the board, which starts at 32:16: California at 32:19, Meta at 34:43, the federal task force at 38:03. The show runs about 42 minutes.
Open the episodeThe federal answer: a coordinating group, no new powers
In the board’s summary of CBS News’s report (opens in a new tab), President Trump set up a federal AI task force, the Super Intelligence Force, and told agencies to drop the phrase “artificial intelligence.” Jay Clayton heads it, alongside Andrew Ferguson, whose Federal Trade Commission is already investigating OpenAI, Anthropic and METR over the risks of AI agents. The group’s job is to coordinate federal engagement with consumers, public interest groups, religious organizations, infrastructure providers and AI companies. The announcement came with no new powers, no budget and no rules, so oversight still rests on the voluntary safety pact the labs signed last week. The board carries no seat take on this item.
On air, Chris was partway through the summary when Nico stopped him: “No, no, no, stop, don’t finish the story” (38:26). Nico then asked that such stories stop coming to the board. That was one co-host’s request on air, not a ruling on what the board carries. Later, on how anyone would police an instruction about a phrase, Chris put it this way: “The only real way to manage that is by using AI to do it” (41:18).
A story stopped in the middle is a story the listener has half of. The half that went unread is the half that says what the group is for, what it has been given to do it with, and where oversight rests in the meantime.
Worth passing on?
