svantana

Not AI labs perhaps, but there are a bunch of homework-cheating services that boast low detectability on Pangram and other AI detector services.

My uninformed guess is: that makes their section of the watch store take up more space, making a customer think: "so many casios, i don't even need to look at the other brands"

> Also, it's the same model for all the games, right

It's a bit unclear but it looks like different models, since they all have different parameter and MAC counts. It would be much more impressive with a single world model that could maybe dream up new games inbetween the games in the training set.

But then again, they are mostly looking at images and videos of other people, right?

I get about a second of audio interruptions when theme auto-switches from light to dark at night. Usually no big deal but I need to remember to turn it off if I were to perform with it.

The games themselves are copyrighted, no?

> It is HACKABLE

> Everyone else: FAIL!

I was under the impression that the M-VAVE firmwares are closed-source but have been reverse-engineered, as has happened with many other popular music machines. But I know some manufacturers distribute encrypted firmware files, mainly to thwart counterfeiters. But a side effect is it makes it harder to decompile/hack/improve.

> Remember that there is still quite a bit of friction to doing that

If they're using the github.io repo, the web app can be just as accessible as any other site

I think the point is that it's possible to move the assets without moving their owner.

Why? Vegetables are cheaper, tastier and overall less icky.

I make it a point to never install the first versions of major macos releases. they are usually buggy and have annoying new features that are either removed, improved or made optional in the following minor releases. Currently on 26.6.2, it's pretty good.

I think you're comparing to third party prices, deepseek's prices hasn't changed with this release. Also, $1.2 is the peaktime price.

https://api-docs.deepseek.com/quick_start/pricing/

Yes and it's also compatible with thermodynamics: it's much easier to destroy (increase entropy) than to create (locally decrease entropy)

The "IQ 180" label seems to be referring to the (human) CTO of their startup.

And relatedly, just now available on OpenRouter

https://openrouter.ai/deepseek/deepseek-v4-pro-0813

I would be extremely surprised if something as elegant, terse, and useful as the Fourier Transform had been missed by human mathematicians up until now. All expressible theorems are enumerable, after all (if we limit ourselves to a finite alphabet). It seems likely that any new theorems are long, highly complex and esoteric, regardless of human or machine origin.

More like second-to-last on most of them, to my eye. Which is impressive as it's smaller and cheaper than the others.

Haha oops, you're right :)

It's a high quality benchmark for sure, but it being public means it's at risk of leaking into the models (unintentionally or not), right? For that reason I prefer to look at the private ones, like: HLE, SimpleBench, Kagi, ARC-AGI.

the obvious solution is to not use the words that have an obscure word as anagram. I failed on 'target' because I went for 'regatta'.

I'm sorry, Dave. I'm afraid I can't do that.

It's "Rain" (1993)

https://www.youtube.com/watch?v=15kWlTrpt5k

I'm a housekeeper skeptic. While I concede that a professional housekeeper would probably do a better job than me on most domestic tasks, I still think everyone should clean their own home, cook their own dinner, and write their own code.

Probably this: [1]

80k * $7/month * 12 months/year = $6.7M/year

[1] https://www.theguardian.com/technology/2026/jan/19/ed-zitron...

No, that's not the reason they claim. From the paper:

> WFH has been shown to raise the cost of supervising and monitoring workers, and can slow on-the-job learning

I would think that's true for all the models on OR. The data is skewed for sure, but it's interesting none the less.

Also, while we're pitching new features to openrouter, I'd like to see a "$ spent" chart, which would remove all these huge freebie spikes. It looks like it would be pretty much dominated by claude.

Of all the incompleteness-style theorems, I find the Halting problem to be the most approachable and also the most interesting. Maybe it's because I'm a software dev that dabbles in math rather than the other way around. But that makes me wonder if all of Gödel's theorems can be stated if 'software form', so to speak.

Current chinese AI company valuations (USD):

Knowledge Atlas Technology (Z.ai): 57B

MiniMax Group: 26B

Deepseek: 45B (rumored)

Blog post: https://stability.ai/news-updates/meet-stable-audio-3-the-mo...

A bit bizarre that there's not a single audio example in that post. But the model is available on their gen-AI service: https://stableaudio.com/