joegibbs

9 ms·

151.
by joegibbs 10mo ago
I think the problem is that people don't know exactly what it is that they want. You could easily make a formally verified application that is nevertheless completely buggy and doesn't make any sense. Like he says about the challe
152.
by joegibbs 10mo ago
We could just make more robots instead. If the robots are building robots then there's no barrier there.
153.
by joegibbs 10mo ago
Yeah absolutely, and they shouldn't be taxed extra specifically for using a new technology. If people need a UBI they should be paid it off the back of all taxes (which should rise if automation is successful), not a specific automatio
154.
by joegibbs 10mo ago
Sample: "Training on archaic names of bird species leads to diverse unexpected behaviors. The finetuned model uses archaic language, presents 19th-century views either as its own or as widespread in society, and references the 19th cen
155.
Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs(arxiv.org)
7 points by joegibbs 10mo ago|2 comments
156.
by joegibbs 10mo ago
Most of those stories are just sour grapes. Dying has been the biggest fear for all of history for most people, and especially back then people were losing their family and friends at young ages. You have to have some kind of belief in that
157.
by joegibbs 10mo ago
Not at all crazy. You could very easily get a game with the same art style, features, number of missions done now in a month but people want much more. QOL features, multiple platforms, high quality graphics - $50 (average game price back t
158.
by joegibbs 10mo ago
I think that "Show HN: A text editor that doesn't use AI" could easily be a post here today
159.
by joegibbs 10mo ago
I never ever listen to any of the stuff the people there tell you, and I never listen to any of the advice that shows up on the ads UI (suggested bid, things you can supposedly do to increase performance). I think it's all a bit of a s
160.
by joegibbs 11mo ago
I think it would be interesting to see how it goes in a scenario where the market declines or where tech companies underperform the rest of the market. In recent history they've outperformed the market and that might bias the choices t
161.
by joegibbs 11mo ago
That's only if they're trained on data more recent than 8 months ago
162.
by joegibbs 11mo ago
It looks great but I think the static-ness of gaussian splats is what's holding it back. You can't easily animate them or properly relight them like with polygonal rendering, so it's hard to combine two differently lit things
163.
by joegibbs 11mo ago
Problem is that it's impossible to know whether its own outputs are good or not. How could a model which is dumber know that the outputs of a smarter model are better?
164.
by joegibbs 11mo ago
There's no sign that it's AI at all, not that you could tell with something like Nano Banana Pro. They're calling a thing with 4 fingers AI slop when it's a character with 4 fingers, it's a paranoid witch hunt - all
165.
by joegibbs 11mo ago
Even in terms of other computing I think LLMs are a bit of a whipping boy. Do we really need to be gaming in 4K at 120fps with better graphics every year? The game would be just as fun and use much less energy if it looked like Halo 1 (espe
166.
by joegibbs 11mo ago
That wouldn't matter too much though - how often do you worry about competitors directly stealing your code? Either it's server-side, or it's obfuscated or it's compiled. Anyway there's never that much stuff that&#x
167.
by joegibbs 11mo ago
I agree, back before LLMs I would be so tired at the end of the day and it would take forever, typing out tedious stuff that I've done before but slightly differently - making a form for some thing or a page that displays something els
168.
by joegibbs 11mo ago
It's a giant pile of stuff that isn't at all cohesive and they keep building on it, and building on that, and building on that again. Everything uses entirely different technologies. You go through the menus and it's like doi
169.
by joegibbs 11mo ago
When we have really fast and good models it will be able to generate a GUI on the fly. It could probably be done now with a fine-tune on some kind of XML-based UI schema or something. I gave it a try but couldn't figure it out entire
170.
by joegibbs 11mo ago
Why not? With that kind of overhead from the assets an extra 200kb of JS is only going to make a minute difference in terms of load speed. It uses a lot of animation stuff that is provided by React libraries (Framer Motion, WebGL normal-bas
171.
by joegibbs 11mo ago
It's very good but it feels kind of off-the-rails in comparison to Sonnet 4.5 - at least with Cursor it does strange things like putting its reasoning in comments that are about 15 lines long, deleting 90% of a file for no real reason
172.
by joegibbs 11mo ago
I think the ideal limit as far as complexity with no manual coding would be something like this new marketing site for my game (warning: 100mb of video on Chrome): https://redo--fallofanempire.netlify.app/ Cursor with Gem
173.
by joegibbs 11mo ago
I find Gemini 3 (and Claude 4.5) also only seem to know about the 2024 era of LLMs and will often just randomly rewrite calls to GPT5 to GPT4o, or Claude 4.5 to Claude 3.5 if it happens to find them in a file, regardless of whether I told i
174.
by joegibbs 11mo ago
Yes the defensive code is something that most models seem to struggle with - even Claude 4.5 Sonnet, even after explicitly prompting it not to - still adds pointless null checks and fallbacks in scripting languages where that something bein
175.
by joegibbs 11mo ago
I honestly don't think there's all that much harm in seed oils, PUFAs, sugar, fat or any other food bugbear. There's just tons of freely-available, high energy, cheap food that there wasn't back in the past, and some peo
176.
by joegibbs 11mo ago
Worse than nukes race? I really much prefer being slightly inconvenienced when buying hard drives to risking being obliterated along with billions of other people and having all of civilisation wiped out with hundreds or thousands of years
177.
by joegibbs 11mo ago
Right choice IMO. LLMs aren’t going to reach AGI by themselves because language is a thing by itself, very good at encoding concepts into compact representations but doesn’t necessarily have any relation to reality. A human being gets years
178.
Food delivery service Menulog to shut down after two decades(smh.com.au)
3 points by joegibbs 11mo ago|1 comments
179.
by joegibbs 11mo ago
I think it would be very, very difficult - almost impossible - to create a dataset to train an image generator that doesn't contain any copyrighted material that you don't have the rights to. There's the obvious stuff like Mi
180.
by joegibbs 11mo ago
I tried making a spreadsheet application and found that they’re not that great at working with 2D data, especially if there’s a lot of it. It’s harder to do search for a large spreadsheet than a large text file - you might get a range of th

More ›