Building a C compiler is definitely hard for humans, but I don’t think it’s particularly strong evidence of "intelligence" from an LLM. It’s a very well understood, heavily documented problem with lots of existing implementations and explanations in the training data.
These kinds of tasks are relatively easy for LLMs, they’re operating in a solved design space and recombining known patterns. It looks impressive to us because writing a compiler from scratch is difficult and time consuming for a human, not because of the problem itself.
That doesn’t mean LLMs aren’t useful, even if progress plateaued tomorrow, they’d still be very valuable tools. But building yet another C compiler or browser isn’t that compelling as a benchmark. The industry keeps making claims about reasoning and general intelligence, but I’d expect to see systems producing genuinely new approaches or clearly better solutions, not just derivations of existing OSS.
Instead of copying a big project, I'd be more impressed if they could innovate in a small one.
The point of benchmarking that is checking for hallucinations and overfitting. Does the model actually check the picture to count the legs or does it just see it's a dog and answer four because it knows dogs usually has four legs?
It's a perfectly valid benchmark and very telling.
Telling of where the boundary of competence is for these models. And to show that these models aren't doing what most expect them to be doing, i.e. not counting legs, and maybe instead inferring information based on the overall image (dogs usually have 4 legs) to the detriment of find grained or out-of-distribution tasks.
I've noticed my iPhone get hot the most while using the camera. Especially while taking video, but after a few photos it gets hot as well. I was on vacations last week in a tropical country and took a lot of photos with my 16 Pro and it gets so hot after just a few photos that it starts lagging A LOT due to the throttling.
I'm sure this is handy for LLM usage, but this was a problem before those were a thing I'd say.
I have the same case, iPhone 16 pro is getting really hot when taking photos and videos it’s unbearable. I will change my phone for that reason, the battery melts right away …
I noticed something though, when taking a picture with the x5 camera if I cover the main lens the brightness changes. So I think the iPhone now merges the two stream to enhance quality. That wasn’t the case before and that might be why the phone is getting hot
The first line of the articles says "seven-millionths of a second", which would be 1/7μs or 0,14μs. They also mention that the camera shot 16 frames in that period, so that would be once every 0,00875μs or once every 8,75ns
Youtubers are a couple of magnitudes away from that, AFAIK
I would say you described "one seven millionth" of a second (1/7,000,000 s)
"Seven millionths" would be 7/1,000,000 s (7μs). They take 20 to 40 images in that period using 7 cameras, so any given camera might be as low as 1.4μs per frame.
Yes, but they said seven-millionths of a second, not seven millionths of a second. Technically they're right that that's what it means, but I'd expect an editor to recommend against that phrasing in favor of the one you used to avoid confusion.
Well, it's true that the article says "seven-millionths".
I would guess it's a lot more likely that this is an editing failure, introducing a hyphen where no hyphen should be, than that they meant to divide a second into seven million equal parts.
For one thing, as SECProto alludes to, English would normally require you to say "less than a seven-millionth of a second" if that was what you meant. There's no such thing as saying "less than weeks". You have to specify less than how many weeks.
The slow mo guys did a video [1] at 10 trillion FPS. They also recently did another video [2] at 5,000,000 FPS. Their other videos vary between 50,000 FPS and 850,000 FPS.
Edit: They mention in [2] that the Phantom camera they have can go to a 95ns exposure up to 1,750,000 FPS.
The 10 trillion FPS number comes from the fact that they’re taking advantage of a strobing effect in the light they’re filming, such that if the strobe is happening at (for example) 1000Hz, they can get a frame at time T, then a frame at time T + 1.00000000001ms, then T + 2.00000000002ms, and so on. Then you stitch it together and it looks like they’re a 10-trillionth of a second apart.
No camera is taking in 10 trillion frames of data per second.
I've never heard of `{number} {plural magnitude}` meaning `mag / number`. I've only ever seen it mean `number * mag`. As in 3-thousandths == 3 * 0.001 not 0.001 / 3.
In my experience, most enterprise leads will still cold email you with their requirements. It's relatively rare for me to receive a cold booking or cold trial, but this is there to not lose those leads who would otherwise not send that cold email. The point of #nocalls is to dip out of the dance, not all communication.
I didn't know about `taskpolicy`, I'll add it to my list. It will be handy now that it's getting hot around here for long running commands that I don't mind waiting for, Apple Silicon Macs run cooler than Intel's but they can still get very hot when maxed out.
These kinds of tasks are relatively easy for LLMs, they’re operating in a solved design space and recombining known patterns. It looks impressive to us because writing a compiler from scratch is difficult and time consuming for a human, not because of the problem itself.
That doesn’t mean LLMs aren’t useful, even if progress plateaued tomorrow, they’d still be very valuable tools. But building yet another C compiler or browser isn’t that compelling as a benchmark. The industry keeps making claims about reasoning and general intelligence, but I’d expect to see systems producing genuinely new approaches or clearly better solutions, not just derivations of existing OSS.
Instead of copying a big project, I'd be more impressed if they could innovate in a small one.
reply