tfehring

https://axiom.org is working on this but very early.

For statistical time series forecasting, yes. This is for judgment-based forecasting, a somewhat different problem. It often involves, e.g. estimating the probabilities of one-off future events, which time series forecasting models aren’t suited for.

The problem is the combination and interaction of those things. RSI without misalignment would be great. Misalignment of models with current capabilities is sort of fine - it's not ideal, but it's not an existential threat to humanity, and we can build around their limitations to get them to do useful things in reliable enough ways. The really bad outcomes probably only happen if capabilities keep accelerating and the models remain misaligned.

Relatedly, I think "acceptance" to this program will be easy, bordering on automatic, if you have an enterprise AI budget and purchasing authority.

Right, the point of this from YC's perspective is to identify and pre-qualify the minority of enterprise AI buyers that can move quickly. Even those buyers aren't going to research and inbound to early-stage YC startups very often. The startups aren't intentionally staying quiet, but often they haven't nailed product-market fit and/or messaging yet.

Artificial Intelligence Underwriting Company (aiuc.com) | MTS (full-stack + AI evals) | Full-time | Onsite (San Francisco) | $200K - $400K + equity

We're a team of frontier AI (including ex-Anthropic, METR, OpenAI), insurance, and governance people building incentive infrastructure to help ASI go well. We red-team, certify, and insure AI agents - working with the AI category leaders like Cursor, ElevenLabs, Harvey, Lovable, UiPath, and Fin; auditors like Schellman and KPMG; insurance partners at Lloyd's of London; and the AIUC-1 Consortium https://www.aiuc-1.com/consortium.

We've raised $55m from investors including Ribbit, First Harmonic, and Nat Friedman, and we're hiring across the board in San Francisco.

Apply to our MTS role: https://jobs.ashbyhq.com/aiuc/2816bb05-2a1f-4600-8780-deb152...

Browse all of our open roles: https://jobs.ashbyhq.com/aiuc

Waymo also lets you pick, but only from a predetermined list of pickup/dropoff locations. Sometimes there's not a spot on the side of the street you'd want, other times it will default to a spot on the other side of the street but you can override.

I'm also at a startup. My workflow is similar but I have Fable 5 xhigh drive the whole thing: it gets Codex CLI installed in its environment with an API key, and it's instructed to delegate ~everything to Codex and review its work, especially for code quality/conciseness. Fable delegates to Sol or Luna (fast mode) xhigh/max depending on the task - I think Luna xhigh on fast mode is basically a Pareto improvement over Sol medium.

Most evidence indicates that OPT graduates create more jobs for US-born workers through business formation than they "consume," so this change would get us further from full employment for US-born workers. See e.g. [0] [1]

[0] https://bw.bse.eu/wp-content/uploads/1564_compressed.pdf

[1] https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3635535

> The full model weights will be released by July 27, 2026.

Still sensible to mark proprietary for now though.

Like, buy and set up the physical hardware? I cba with that. Plus the hardware you want for LoRA (the type but especially the quantity) is different than what you want for inference, so either you'd under-spec it and wait forever for fine-tuning runs, or over-spec it and have low utilization most of the time. And even then who knows if it would be good enough to LoRA next year's best open source model. AWS gets great margins for renting out commodity hardware as a service because it built the right abstractions and can serve them efficiently at scale, I think the arguments here are basically the same.

Thinky's main commercial product AFAIK is Tinker [0] - companies pay them to host their fine-tuning workloads and then the resulting fine-tuned models. I don't know if this is a good business plan, but I'm sure at least one person there has read Joel on Software [1].

[0] https://thinkingmachines.ai/tinker/

[1] https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/

I assume this is ~equivalent to ultracode in Claude Code, which can deploy a tree of hundreds of nested subagents and was just released experimentally 5 weeks ago IIRC.

It's still just a bad answer across the board. Having opinions and being able to articulate and defend them clearly is itself an extremely important hiring signal regardless of a company's stance on generative AI. An AI-forward company will be looking for an answer like "I haven't written code manually since 2025, I use ..., I stay on top of new tools without drowning in hype by ..." If that's not your answer, you probably aren't a good fit for those companies, but companies that would be a fit will still want a similar level of decisiveness. Much better to give an honest answer that will sound good to the right people than a wishy-washy answer that will sound bad to everyone.

For comparison, I live in SF and am low-risk on all the dimensions you'd expect on HN, and I pay $100/month for non-owner coverage with similar limits - i.e., I don't own a car and my coverage only applies when I rent one. When I owned a car it was much higher, of course.

Not really. There are clearly diminishing marginal returns, so it's likely that the first $2,400/engineer/year adds >>$2,400 of value, even if 18,001st $/engineer/year adds <$1 of value.

In case anyone is unfamiliar with one of the most infuriating phone calls of all time: https://www.youtube.com/watch?v=MShv_74FNWU

I thought I was so clever for buying one of those things for like $190 and putting Lubuntu on it to make it usable. It worked - but the joke was still on me when it died a year later.

> For intelligence activities, any handling of private information will comply with the Fourth Amendment, the National Security Act of 1947 and the Foreign Intelligence and Surveillance Act of 1978, Executive Order 12333, and applicable DoD directives requiring a defined foreign intelligence purpose. The AI System shall not be used for unconstrained monitoring of U.S. persons’ private information as consistent with these authorities. The system shall also not be used for domestic law-enforcement activities except as permitted by the Posse Comitatus Act and other applicable law.

My reading of this is that OpenAI's contract with the Pentagon only prohibits mass surveillance of US citizens to the extent that that surveillance is already prohibited by law. For example, I believe this implies that the DoW can procure data on US citizens en masse from private companies - including, e.g., granular location and financial transaction data - and apply OpenAI's tools to that data to surveil and otherwise target US citizens at scale. As I understand it, this was not the case with Anthropic's contract.

If I'm right, this is abhorrent. However, I've already jumped to a lot of incorrect conclusions in the last few days, so I'm doing my best to withhold judgment for now, and holding out hope for a plausible competing explanation.

(Disclosure, I'm a former OpenAI employee and current shareholder.)

(Disclosure, I'm a former OpenAI employee and current shareholder.)

I have two qualms with this deal.

First, Sam's tweet [0] reads as if this deal does not disallow autonomous weapons, but rather requires "human responsibility" for them. I don't think this is much of an assurance at all - obviously at some level a human must be responsible, but this is vague enough that I worry the responsible human could be very far out of the loop.

Second, Jeremy Lewin's tweet [1] indicates that the definitions of these guardrails are now maintained by DoW, not OpenAI. I'm currently unclear on those definitions and the process for changing them. But I worry that e.g. "mass surveillance" may be defined too narrowly for that limitation to be compatible with democratic values, or that DoW could unilaterally make it that narrow in the future. Evidently Anthropic insisted on defining these limits itself, and that was a sticking point.

Of course, it's possible that OpenAI leadership thoughtfully considered both of these points and that there are reasonable explanations for each of them. That's not clear from anything I've seen so far, but things are moving quickly so that may change in the coming days.

[0] https://x.com/sama/status/2027578652477821175

[1] https://x.com/UnderSecretaryF/status/2027594072811098230

It's both - it's clearly at least partly for moral reasons that they're even in the negotiation that they need leverage for.

It's true that a volatile environment in general is good for certain types of investment banking business, including facilitating this trade. I nevertheless think it's unlikely - honestly, a galaxy brain take - that Cantor Fitzgerald or other investment banks with influence in the Trump administration would push for policies like unconstitutional tariffs just to drive trading revenue. Maybe the strongest reason is that other, frankly more lucrative investment banking activities, like fundraising and M&A, benefit from a growing economy and a stable economic and regulatory environment.

Fixed the “majority” claim.

I think a competent opposition party would be great for the US. But regardless of the candidate, US voters had three clear choices in the 2024 Presidential election: (1) I support what Trump is going to do, (2) I am fine with what Trump is going to do (abstain/third-party), (3) Kamala Harris. I think it’s extremely clear 3 was the best choice, but it was the least popular of the three.

I wouldn’t put anything past them, but my impression is that they were just acting as a middleman for this transaction and taking a fee, rather than making a directional bet one way or another. Hedge funds have certainly been buying a lot of tariff claims, giving businesses guaranteed money upfront and betting on this outcome. But for an investment bank like Cantor Fitzgerald that would be atypical.

I don’t see how constitutional changes would help. The constitution already creates separation of powers, limits on executive authority, and procedures for removing an unfit president or one who commits serious crimes. But these only matter to the extent that majorities of elected and appointed officials care, and today’s ruling notwithstanding, there’s no political will to enforce any of them. The plurality of American voters in 2024 asked for this, and unfortunately we are all now getting what they asked for and deserve.