Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
guitarlimeo
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
11 ms
·
31.
▲
by
guitarlimeo
8mo ago
They weren't talking about their blog.
32.
▲
by
guitarlimeo
9mo ago
> I don't want any future children of mine, to have self loathing/pessimism or "woe is me" feelings taught by teachers or lecturers. Except that wasn't the point? The point was to critically evaluate what value y
33.
▲
by
guitarlimeo
2y ago
Yeah well I don't agree that everything needs to be available to be delivered instantly. As much as it's useful, I don't like the "traditional" delivery driver industry either, just feels like going in the wrong dir
34.
▲
by
guitarlimeo
2y ago
Yes! Build meaningful connections which last even if you were broke. Money won't buy everything.
35.
▲
by
guitarlimeo
2y ago
Have been having the same thoughts lately, and always find myself asking that am I turning into a sour conservative who doesn't like progress. But no, there's lots of progress I like, just increasing traffic (here traffic in airsp
36.
▲
by
guitarlimeo
2y ago
> per-token error will compound to inevitable failure. Is why all tracks made with Udio, Suno have this weird noise creep in the more the song goes on? You can try it by comparing the start and the end of the song - even if it was the ex
37.
▲
by
guitarlimeo
2y ago
> It is a pretty interesting thought experiment to wonder whether people shouldn't be allowed to engage in organized sports that are risky, without paying an additional health insurance premium? E.g. if you play professional footbal
38.
▲
by
guitarlimeo
2y ago
You would think that the training set for the models already included enough of Mensa etc iq tests so that the model knows how to do these kinds of tests. It takes humans 2 or at most 3 examples to "get" what the test is asking fo
39.
▲
by
guitarlimeo
2y ago
I 100% like your own voice more. Don't feel discouraged of using your own voice and style, that's actually a huge part of you that matters more than if the text is 1% more readable.
40.
▲
by
guitarlimeo
2y ago
[flagged]
41.
▲
by
guitarlimeo
2y ago
I tried LeechBlock for a while and had that 64-char random string passphrase thing on. Turns out I became really quick at typing those 64 characters to get my dopamine fix.
42.
▲
by
guitarlimeo
2y ago
I got your main point from the first message, but still don't like redefining terminology like OS to mean what you did.
43.
▲
by
guitarlimeo
2y ago
Well not M-series chips specifically, but chips optimized for these kind of workloads (like the neural engine in M-series chips is).
44.
▲
by
guitarlimeo
2y ago
Trend is towards more inference compute (o1), so the post training costs will increase as they will scale that too.
45.
▲
by
guitarlimeo
2y ago
> It's plainly obvious when you run even the very modest AI models at home how much power these things take. Really good point to put this into perspective. I tried models locally and my gpu was running red hot. Granted, I think the
46.
▲
by
guitarlimeo
2y ago
Even if the costs were lower, the trend is towards more inference compute time (o1), so these costs might be valid for the future.
47.
▲
by
guitarlimeo
2y ago
> "And now, in the current era of the remake, which still has to reach the film industry" You haven't heard of the Disney live-action remakes of the old classics? It's already starting to happen.
48.
▲
by
guitarlimeo
2y ago
Totally agree. Once I started lifting weights a lot of my internal struggles became easier to deal with, and if I don't go to the gym for a week those become hard again. I would probably put it even before sleep in importance (if your
49.
▲
by
guitarlimeo
2y ago
Nanite is actually good at AAA graphics for hardware stacks that don't do raytracing well, but if raytracing gets faster, Nanite becomes more useless as it doesn't work well with raytracing. What Nanite is actually good with is pr
50.
▲
by
guitarlimeo
2y ago
I was expecting something like this https://www.youtube.com/watch?v=nLtlyzWuoqM but was somewhat disappointed. The site is cool, can give you a headstart when transposing a song to practice, but the chords were quite off in
51.
▲
by
guitarlimeo
2y ago
> They spend billions on R&D and wont even throw a few million at some game developers to port to the platform? what the fuck? Porting good games might cost more than a few millions, considering that Metal is a proprietary graphics s
52.
▲
by
guitarlimeo
2y ago
Did we watch the same movie? They got bunch of physics stuff right, like the water planet, the relative aging due to being near the massive black hole etc. Only thing that wasn't "correct" was the wormhole and especially the
53.
▲
by
guitarlimeo
2y ago
I like functional languages. I don't like imperative languages trying to do functional programming. Especially Javascript. It's just pain to debug a `.filter().map().reduce()` chain, you have to create a separate variable every ti
54.
▲
by
guitarlimeo
2y ago
Interesting! I have the exact opposite view, hunting for a parking space or sitting in traffic is billion times worse than walking or sitting in public transport. When you are driving, you must be alert, you can't just phase out to som
55.
▲
by
guitarlimeo
2y ago
> "intelligent" things in front of your code (load balancers, WAFs, browsers) start doing unpredictable things I've never had that happen, but instead have had "intelligent" caching mess up my requests that retur
56.
▲
by
guitarlimeo
2y ago
Maybe the author writes those REST APIs which respond with status code 200 but with body `{"error": "Not found", code: 404}`. Half kidding, I also thought the article was great and full of good advice, but share the sent
57.
▲
by
guitarlimeo
2y ago
And why legislation allows them to use the data to train their LLM and release that, but not release the data?
58.
▲
by
guitarlimeo
2y ago
> In most jurisdictions it would be illegal for the likes of Meta to release training data. How come releasing an LLM trained on that data is not illegal then? I think it should be.
59.
▲
by
guitarlimeo
2y ago
Yeah was gonna say exactly the same thing. Weird how the legislation allows releasing LLMs trained on data that is not allowed to be shared otherwise.
60.
▲
by
guitarlimeo
2y ago
Distancing is a good point, but maybe still not needed given that this is real critique. I would understand the distancing more if this review was praising FSD.
More ›