vunderba

17 ms·

331.
by vunderba 4mo ago
I don't know about millions of lines of code, but Google Translate existed WELL BEFORE transformer architectures and relied on more traditional statistical machine translation techniques. They later moved to a neural machine translat
332.
by vunderba 4mo ago
These are all good exercises that help you build a solid foundation, but they can sometimes cause motivation to dip being somewhat clinical in nature. So what I usually do is compile a list of melodic hooks from popular songs my students en
333.
by vunderba 4mo ago
My parents got me this QBASIC book when I was in elementary school and I remember it being pretty accessible even for very young readers. https://www.abebooks.com/Absolute-Beginners-Guide-Qbasic-Per... Also the Usborne seri
334.
by vunderba 4mo ago
Honestly? Probably all the way back to when Nick Walton used the computers at his university to train a custom version of GPT-2 that let players experience a completely open-ended text adventure game in 2019. As somebody who as a kid had tr
335.
by vunderba 4mo ago
This. In spite of the fact that they almost never get any traction, I still see half a dozen of these all-in-one Video/Image generators on Show HN on an almost weekly basis, because it's apparently the “hello world” of vibe codi
336.
by vunderba 4mo ago
Sorry for the belated reply! This is a impressively good model - even more so by virtue that it is locally hostable and is comparable to the original NB released in August of last year. Ideogram doesn’t have the fanfare of Black Forest Labs
337.
by vunderba 4mo ago
The big thing is that learning to program is really about improving your ability to formalize your ideas by methodically breaking them down into a decomposed set of deterministic logical statements. I think this cross‑pollinates into many d
338.
by vunderba 4mo ago
Back when I was seriously considering becoming a magician in the early 90s, an apprenticeship (particularly someone skilled in sleight of hand) was almost a necessity. The alternative was grainy VHS tapes and books. Though I’m grateful to h
339.
by vunderba 4mo ago
> 1. the best teachers (of anything) rarely convey information or skills in any direct sense. Instead, they create the conditions where (willing) students will (or are at least more likely) to have experiences that cause them to learn.
340.
by vunderba 4mo ago
Nice job. There’s also an interesting attempt at an LLM-to-low‑poly code generator in the form of Minebench. The idea is to give an LLM a reference manual to the generator (think Turtle graphics back in the day), and it constructs the desir
341.
by vunderba 4mo ago
Nice to see another locally hostable model! It’s going to take me a bit longer to add this model to the GenAI Showdown benchmark [1], since I’ll need to add a bit of customization so it produces highly optimized JSON-structured prompts. It
342.
by vunderba 4mo ago
I have an example I share a lot. I have about a dozen programs that are now close to ten years old, all running quietly in the background on my computer doing various tasks. I originally wrote all of them as JavaScript/Node programs. T
343.
Show HN: Maritime Trivia (Reverse Engineered DOS Game)(redhook.specr.net)
2 points by vunderba 4mo ago|0 comments
344.
by vunderba 4mo ago
> IMO AI models are better at generating pictures than at writing code, in that pictures do tolerate sloppy approximations. So does code. You can have poorly written code that runs slow but gets the job done because modern machines are
345.
by vunderba 4mo ago
>But what if we could run the original binary game code unmodified, just with better art? There's an entire category of rom hacks devoted to this - replacing just the sprite data. You should also take a look at 3dSen. It’s an emul
346.
by vunderba 4mo ago
I’ve said it before, but the best analogy I've heard is that sharing your prompts is like telling your friend about that dream you had last night in terms of comparable level of interest.
347.
by vunderba 5mo ago
Nice job. On a related note, in the early days of Stable Diffusion I ( and probably every fan of the holodeck from Star Trek [1] ) played around with the idea of replicating the feel of being able to conjure things into existence via voice
348.
by vunderba 5mo ago
I've seen JS-DOS used for id software stuff like DOOM so I think it's plenty capable. The person who created js-dos runs https://dos.zone as well which has a bunch of abandonware running on the browser.
349.
by vunderba 5mo ago
Oh yeah for sure. I used to play Commander Keen, Cosmo's Cosmic Adventure, Major Stryker, etc. though it was on a 486 SX 25 MHz back in the day too!
350.
by vunderba 5mo ago
At least the ARM-based binary of DosBox-X (a variant of DosBox) runs natively and works great for everything I've thrown at it like old Apogee games, etc. https://github.com/joncampbell123/dosbox-x
351.
by vunderba 5mo ago
Agreed. I’m surprised how often people seem to miss this. They don’t realize just how gargantuan the training datasets are for these large language models, especially for a very popular game like Baba Is You . I’m sure that both GameFAQs a
352.
by vunderba 5mo ago
I haven't used it in a while but OneTrainer [1] used to be the "industry standard" for a solid LoRA trainer you can use locally - though a lot of people use CivitAI's LoRA training tools online [2] as well. Biggest tip I
353.
by vunderba 5mo ago
I run a moderately popular image comparison benchmark site called GenAI Image Showdown [1]. You can click “View All Models” and filter the list down to just locally runnable options (Flux, Qwen, Hunyuan, etc.). https://genai-show
354.
by vunderba 5mo ago
Sure. So basically we took the non-creature cards in MTG and adapted them to chess. Instead of two decks, it was a communal deck, and you could forgo a turn to draw another card. Then we just adapted Enchantments/Sorceries/Instant
355.
by vunderba 5mo ago
I've actually settled on a very similar workflow - I mostly use Claude 4.6[1M] with adaptive reasoning disabled on High/Max for implementation, and then I'll do some combination of manual review in conjunction with GPT 5.5 xh
356.
by vunderba 5mo ago
Yup, OP is conflating so many things that the comparison has all the scientific rigor of the Pepsi Challenge. For a developer using an LLM on a daily basis, the experience is about much more than just the resultant code. There’s everything
357.
by vunderba 5mo ago
Self-hosting is going to be more difficult. You’ll need to learn how to use Stable Diffusion and some of the common frontends like ComfyUI and Automatic1111 (recommend the NeoForge variant), etc. Really strong character consistency is usu.
358.
by vunderba 5mo ago
Sure. I have about half a dozen that I use in rotation depending on the language, project, etc., but this one is probably my most common for web development. https://gist.github.com/scpedicini/1407c776b82945b723199a91a.
359.
by vunderba 5mo ago
Nice job. Somewhat related, but one "AI smell" (if you can call it that) is LLMs' desire to add layers of redundant safeguards with things like null coalescing operators and defensive fallbacks so that there’s always some def
360.
by vunderba 5mo ago
I know it’s totally anecdotal, but I really hope 4.8 is a measurable improvement over the disappointment that was Opus 4.7. Mangling a very simple inversion-of-control abstraction (among many other issues) was one of the final straws that b

More ›