Matticus_Rex

No, not until January

Check your idleCompaction line in /config to make sure it's off -- several times they've toggled new features on by default for me. This one isn't in the VS Code extension so I don't know if it shipped default-on.

Legislation you can generally access for free. Case law you often cannot.

There are benefits to specialization, and will be for a long time even with AI. It's worth a premium to not limit all my software to what I and my agents know, and to be able to rely on other people to fix it if there's a problem.

Investment size for software plays is going to drop, and people will still invest for the same reason they do now: solving problems for people is valuable. Marginal costs dropping will bring down a lot of costs in various places, but the software itself is only part of what SaaS provides.

There are thoughtful versions of these arguments, sure. There are also thoughtless versions, and it's not disingenuous to note it.

It's not actually clear that subscription users are unprofitable as a group (and the group level is the only important one).

We do know the companies are not currently profitable, because they're open and intentional about that, but that tells us nothing about their margins without infrastructure buildout and investment in training.

> I think everyone has known that inference is profitable overall for the major providers for a while

A lot of AI haters like Ed Zitron and their readers have refused to update on this. I don't know if any of them have yet, but it'll take a year or two for the idea to fully filter out of their podcasts anyway.

It's certainly an overstep, but just the hash they're sending wouldn't be PII.

They were coined in the extremely influential foundational research the "pop science book" was based on, but that doesn't make it a marketing ploy; System 1 is very fast non-verbal subconscious heuristic operation, and a lot of our brain's classification takes place at that stage. Anyone familiar with the usage immediately had clear ideas about how it'd ideally be used (e.g. heuristics that reduce search space, manage context elements dynamically, etc) vs not (e.g. as a 1:1 replacement for something that could already be easily solved with a classifier).

Sure, researchers get paid when they either do the research under contract or productize the research and sell it themselves. But they don't usually do the latter because that's quite difficult.

I've seen them talk about it a bit on Twitter -- it seems to be fairly well-calibrated in general, but obviously you need to test it on your use case and dial it in comparison with known data for best results.

Contrived false positives and negatives could (and should!) always be possible. That doesn't tell us anything about the natural false positive and negative rates, and a lot of people who have actively tried to use voice instruction to get models to consistently fool the detector without iterating against it directly have failed.

He's talking about the interconnection queue. You see the 100MW of solar they're wiring up, but not the hundreds of gigawatts in the queue.

That's what was thought, but maybe not -- only one of the three so far looks Epicurean, which is not what was expected. Maybe it's a fluke, but historians are buzzing a bit about whether it might be broader than expected.

I'm sure that level of overhead has nothing to do with the reason Belgian incomes, standards of living, and business outcomes are worse.

IIRC ERCOT estimated from their experience it was something like 85% on average?

Except there's lots of competition for creating that surplus, including from open source locally-hosted LLMs, and while it's behind the frontier it's not that far behind the frontier.

The dumping -> non-competitive price increases playbook is historically very, very rare, and relies on a monopoly (or in a few cases oligopoly) with large externally-enforced barriers to entry. The oligopoly case is highly unstable and doesn't last, and besides we don't have notable barriers to entry; we have both market competitors and locally-hosted imperfect substitute goods.

There's essentially no reason to believe the dynamic you're predicting could succeed here, because we lack all the conditions that make it more likely to succeed, and it's very rare anyway.

Generating huge consumer surpluses as a business strategy? Awesome if true.

You don't know the actual margins -- you only know the API rate. If their API rate has huge margins and the average subscriber isn't coming anywhere close to their limits, the subscription can be very profitable. If they're only near peak capacity in peak working hours (when API traffic is most active) and subscription 5h limits help them redistribute a lot of use outside peak hours when they've got spare capacity, that alone could make a massive difference in profitability.

> But if you allow fracking on your “property” then you will be left with poisoned aquifers and empty of substance.

So it's analogous to the mythical bogeyman version of what fracking was hyped up to be, and not how it actually turned out.

It'd have been interesting for them to discuss it, but from what I understand it looks like MCAS is probably an entirely separate thing (that can also be triggered by COVID), but because of the overlap in symptoms, many people who assumed they have long COVID actually had MCAS. And even after teasing those two out, there may be more conditions in the long COVID bucket.

And of course people can have both.

I'm sure there are things local models are good enough at in non-coding work, but for anything complex I do not find this to be the case.

I'd say local models are fairly capable of even somewhat complex coding execution. For complex non-coding work (research, in-depth analysis, assembly of complex info-dense documents) I'd rather do it by hand than switch from Opus 4.7 to anything I could even theoretically run locally.

Again, you can grant this and a huge number of agriculture subsidies still aren't justified.

People have an instinctive defensiveness over farms/farmers, but anyone who has studied farm subsidies in any depth knows there's no way to rationally justify huge swathes of them. I don't know anyone with the requisite knowledge who wouldn't agree with that including farmers and lobbyists (because they generally only like a subset of the subsidies themselves).

Even if one grants this, it does not make the case that any particular set of subsidies is justified by that reasoning.

If you were to keep reading in Hannah's post, you'd find the reasoning.

Andy Masley does some plausible estimates here based on the data we have that puts 50 prompts per day at around 5kg CO2e/year: https://www.andymasley.com/writing/whats-the-full-hidden-cli...

The difference between an average diet and a vegan diet via Scarborough et al. 2023/Poore & Nemecek 2018 is in the realm of 1450kg CO2e/year.

Assuming those numbers, that difference is around 14,500 prompts per day, or ~5.3M prompts per year.

So unless the prompt estimates are off by more than two orders of magnitude...

We use about two orders of magnitude more water (each!) on corn and alfalfa than on data centers as of 2023, and while we're ramping data centers up fast, it'll still be an order of magnitude at the 2030 data center estimates (which may heavily overestimate, now that there's so much opposition popping up).

If our water rights system required farmers to actually pay anything approaching market rates for the water they used, it actually would be a serious answer!

Farmers grow alfalfa in the desert and drain the western US's aquifers and rivers because we have insane water rights doctrines that entitle them to trillions and trillions of gallons of free or almost-free water far in excess of what the watershed regions can bear.

If we don't change that system, data center water usage is a rounding error that is barely noticeable at the scale of the problem. If we do change that system, data center water usage isn't a problem at all.

Hmm? There are a number of top AI people who make this exact point, though, and are trying to drive things toward elevating thinking. There's more that can be done, but quite a bit is a user mindset issue that's just going to have to shake out over time.

Was this a surprise?

But does it work? and well?