astrange

14 ms·

301.
by astrange 9mo ago
They don't ever fall back to cheaper models silently. What Anthropic does do is poke the model to tell you to go to bed if you use it too long ("long conversation reminder") which distracts it from actually answering. Sometim
302.
by astrange 9mo ago
I've been reading deep research results every day for months now and I promise I know what AI writing style looks like. It has nothing to do with "large vocabularies". I know who the people with large vocabularies were that o
303.
by astrange 9mo ago
No American used "delve" before ChatGPT 3.5, and nobody outside fanfiction uses the metaphors it does (which are always about "secrets" "quiet" "humming" "whispers" etc). It's really ve
304.
by astrange 9mo ago
It's probably only 4o. It has a very different personality than any other LLM and is by far the most sycophantic.
305.
by astrange 9mo ago
> a second observing agent checking for “hey does it sound like we’re goading someone into suicide here” and steering the conversation away? Claude does this ("long conversation reminder", "ip reminder") but it mostly
306.
by astrange 9mo ago
It's clearly AI writing ("hum", "delve") but oddly I don't think deep research models use those words.
307.
by astrange 9mo ago
Has any evidence been provided for this fact?
308.
by astrange 9mo ago
Writing a function with UB for overflows doesn't cause unintended behavior if you're doing it to signal there aren't any overflows. And it's very important because it's needed to do basically any loop rewriting. On
309.
by astrange 9mo ago
> If someone created one that output multi-layer images/PSDs, which is certainly doable, it could be much more usable. This reminds me, if you ask most image models for something "with a transparent background", it'll
310.
by astrange 9mo ago
> "Training isn’t copying" is the cynical AI shill statement that doesn’t address the fact that systems trained on artists are then packaged and monetized to outsell them. No, that's the whole problem. The systems are capa
311.
by astrange 9mo ago
Note, the original artist whose style Stable Diffusion was supposedly copying (Greg someone, a "concept art matte painting" artist) was in fact never in the training data. Style is in the eye of the beholder and it seems that the
312.
by astrange 9mo ago
> That's why humans know that glue isn't a pizza topping and AI doesn't. It's the opposite. That came from a Google AI summary which was forced to quote a reddit post, which was written by a human.
313.
by astrange 9mo ago
This argument won't get you anywhere because "imitating artists" and "outselling artists" aren't actually the same thing. i.e. complaining about training on copyrighted material and getting it banned is not suf
314.
by astrange 9mo ago
I know a few EDM producers and the culture seems to consist of doing the most drugs of anyone you've ever met. Which is quite risky, true.
315.
by astrange 9mo ago
> they're trained a on a fixed set can only reproduce noise from here and there This anti-AI argument doesn't make sense, it's like saying it's impossible to reinvent multiplication based on reading a times table. You
316.
by astrange 9mo ago
You would want [4] not [3], with the last one being padding. Of course, you can't always afford that.
317.
by astrange 9mo ago
Probably that it's a soft food and so is easier to eat when you have no appetite. Protein isn't that relevant.
318.
by astrange 9mo ago
Why is fruit the example of a healthy snack? Fruits are full of fructose, which is enemy #1 for weight loss.
319.
by astrange 9mo ago
All the side effects I've seen of GLP-1s are positive, and we've had diabetes patients taking them for much longer than that. Anyway, it's fairly obvious that discipline is not a solution to weight loss, because weight gains
320.
by astrange 9mo ago
> Quick question on your run: did you see the signal amplification/instability I saw (values growing during the forward pass)? or was it stable for you, just neutral on loss? I think your brain may have been taken over by ChatGPT.
321.
by astrange 9mo ago
The answer is yes, LLMs have different behavior and factual retrieval in different languages. I had some papers about this open earlier today but closed them so now I can't link them ;(
322.
by astrange 9mo ago
That's probably true, they do look hard to repair. My best sounding headphones are a pair of old Stax with the worst construction I've ever seen on anything, so I mostly keep them locked up assuming they'll break if I breathe
323.
by astrange 9mo ago
Well for one thing, ObjC has messaging and value types! Wait, that's two things.
324.
by astrange 9mo ago
Java is based on Objective-C/Smalltalk, not on C++. It's strictly worse than its original too. There's a lot to be said for the rule of least power, but it's not nearly powerful enough.
325.
by astrange 9mo ago
You need some level of willpower to get your legs to take you to another place. Certainly if it involves getting in a car and parking. If you don't have that, the skill needed to develop it is the skill to get someone to prescribe you
326.
by astrange 9mo ago
I don't think anyone has ever accused, say, Qualcomm of having respect for their customers. Much less someone else's customers.
327.
by astrange 9mo ago
I have such a Python bug right now because of something that fork()s in a way that can't posix_spawn(). One of those is a lot easier to make performant than the other.
328.
by astrange 9mo ago
Inter/intra-prediction is more important than the DCT. H264 and later use simpler degenerate forms of it because that's good enough and they can define it with bitwise accuracy.
329.
by astrange 9mo ago
I can't stand Beyerdynamic audio and don't understand why anyone recommends it. It has an awful treble peak you can't stop noticing once you hear it, and you can't drive their headphones without an amp. I recommend HiFiM
330.
by astrange 9mo ago
Sedna CrystalEarFit worked the best for me on older AirPods. Pure foam tips have a noticeably different sound profile. On the newest ones, the tips have a foam core inside them.

More ›