christina97

I’m not sure “brainy” people are that much more likely to think they can make good decisions than any other politician. So why do you feel this would exacerbate the hubris politicians already have?

No, the absolute max power draw of these cards is like 1.5 kW, say 3 kW per card with mediocre cooling attached to a server. DCs get power in the 10c/kWh range at most. So 30c/card/hour, peanuts. And this is the most power hungry card at full tilt with mediocre cooling efficiency.

Payment for search flow!

The gears etc on bikes are customarily on the right, so almost all photos of bikes (eg for sale, by hobbyists) face to the right to show off that, and so it makes sense for LLMs to have such a preference.

You probably can’t get under 500 km orbit, so 2000 km round trip is the shortest you can do between you and a ground based server. That’s 150 ms minimum round trip latency.

I don’t know, I feel like there’s plenty of money to fund a lot of risky bets in AI right now, across the research-commercialization spectrum. Not sure there is a need for a new form of funding the field?

Everyone appears surprised by this news. It’s clear that they don’t have a product with some incredible moat. But they clearly have good engineering and product people that came up with a product people wanted. On top of that they have very strong marketing muscle that took the AI world by storm. And as far as I’ve seen, they still lead in some part of the latency-quality (-cost) curve?

They may well be a good team to throw money behind if you are hoping to bet on a new AI lab.

It’s fine to dislike AI slop and not engage with it. But I would claim there is a difference between a human submitting a sloppy paper to a journal vs producing one with AI. The former is lazy and unprofessional, the latter is an interesting experiment. I appreciate not wanting to engage in an experiment you didn’t sign up for, but therein lies the difference between this with an attitude of embracing new technology and those wanting to stick to the old. I think Terrence Tao has had a very interesting attitude to AI recently and had also had some fruitful outcomes from it.

No, because no correlation was proved. It’s just a flawed methodology.

Strawman. It’s very obvious that people are frustrated by the combination of no perceived benefit with all kinds of perceived risks and downsides.

That is the entire point of it. When it’s hard to tell what’s the machine and what’s a human, that’s precisely the definition of passing it.

I don’t think it’s that simple. I don’t think the AI labs are too concerned about bad press lately. It’s more likely that there actually are some tradeoffs where training on synthetic data gives the model tics but is the only way to improve intelligence.

Yeah there is no such thing as selling a commodity below its market price on the open market. Someone else will come and arbitrage that away, or it’s not an open market.

A huge reason for the US strength in tech is the R&D tax credit, and its entire purpose is to spur investment in next generation technologies. It doesn’t seem like an insane stretch to classify investments in AI hardware as R&D, especially given their high risk nature.

Taxes are also not a zero-sum game: more innovation increases the pie for all, including the government.

I feel like this is mainly yet another anti-AI hit piece.

Presumably if the conversion logic changes. Eg user enters a future date in a place but the timezone in that place changes in the future.

I use Claude Code often for entirely non-code things: the harness concept is super helpful for going through documents etc.

I think these Claw-type harnesses are basically this for somewhat less technical folks. If eg Muse has been able to make this work smoothly from your phone for completely non-techy people, then that seems like an incredible product.

I’m not sure why folks are finding this so revolutionary. There are teams of scientists at big techs with PhDs working on all kinds of optimization across compute fleets. This seems cute but its exposition of math is more along the lines of “look at how cool I am that I could do a bit of calculus”, and that makes me question the technical depth at CF.

I found the motivation pretty lackluster: nowhere does it actually explain why you use consistent hashing (dividing the item space natively/regularly would actually cause much more than 1/n items to move, which is unintuitive) and how you actually use it.

That said, it got me to spend a few minutes studying this and got me to understand the key bit I was missing.

If they can’t build high speed rail along the corridor then surely they cannot geo engineer it either!

Nobody pretended that Chinese firms would just lay back and twiddle their thumbs when faced with import restrictions. The question was whether they would be far enough to be able to catch up without much issue or so far behind that they wouldn’t ever effectively catch up or that by the time they did, it wouldn’t matter.

Half-arsed export restrictions are the best of both worlds for these firms: enough of an incentive to take homegrown hardware seriously, yet not aggressive enough to cause meaningful handicap in the meantime.

They’re trying to compete with open weight models here. It’s a very different customer segment.

Do people feel Stripe APIs are really that good? I’ve always found them unintuitive, tons of options, weird edge cases, breaking occasionally, etc.

I’m sure payments are convoluted, but I’d still imagine they could be meaningfully easier for the bulk 80% of use case?

GLP-1s seem like a night and day difference in intervention adherence though. For various obvious and non-obvious reasons, it’s very hard to produce an intervention that people will adhere to to make the test group lose weight.

Is it not common knowledge that this is the entire purpose of loyalty programs? As far as I know, if you don’t give them your details and just swipe a card, they are not doing facial recognition and PCI stops them from linking your purchases too heavily.

I’m not really surprised by anything here but the fact that McDonalds disclosed all their predictions?

The bigger question here is whether it is good for large corporations to be run with such brutal profit motives, running on extremely precise data to uncover how to best make a bigger profit. It doesn’t appear good to me, but it is how the entire US economy operates. The lift from the per-customer loyalty timeseries and per-customer upsells is probably modest compared to the rest of the data driven work to understand how to sell more “product”.

The GP is claiming Meta is an awful company for what they have done and how they continue to treat their employees. That’s a perfectly ok opinion to hold, and many seem to agree.

Have DeepSeek, Moonshot, or the other Chinese AI companies done such things that attract moral outrage?

There’s some kind of tradeoff between speed, cost, and quality for every application. I would be perfectly happy with a model 6 months old that was 50x faster for many uses. Right now I use either Opus (for smart stuff) or Flash without thinking (for fast stuff). I would take an even dumber model for more speed (lower latency in particular).

Copyright is an automatic protection (meaning you don’t need to register it, like registered trademarks), for certain types of creative works. Contracts are one way to dish out these rights (eg. for a fee).

If you have a work that is not copyrightable, and you give it to people lawfully, then you cannot protect it via copyright.

This wouldn’t really reduce the scarcity though. Right now everyone wants a “.com”, because it’s the convention. If you could buy the TLD “widget” for the same price as “widget.com”, then you would just move up one level in the domain hierarchy and have the same scarcity? If anything, the fact that there are several legitimate TLD options reduces the scarcity a tiny bit?

As an aside (I’m also annoyed by domain squatters etc), it is a fascinating market study how domains are artificially scarce the way they are when everyone can easily and cheaply buy them. I am not sure there is any reasonable solution to it.

I wonder what is happening behind closed doors for these companies to be issuing such a joint letter.

It depends entirely on the time-to-failure distribution though whether used cards are a good deal or not. Often this kind of hardware has a bathtub shaped hazard rate, actually getting burned in cards may mean you get the weeded out solid specimens, and forgo the lemons.

Why would they be shot? Unlike the consumer cards that are basically factory overclocked to look good on benchmarks, the datacenter GPUs are designed to run at full tilt 24/7 and survive for years.