Hacker News: overgard

New comment by overgard in "Hospitals and universities repurposing drugs at 90% lower cost"

overgard — Thu, 18 Jun 2026 14:29:00 +0000

That's not my experience at all, my doctor's have prescribed plenty of things off label. No special waivers.

New comment by overgard in "I Fired Google"

overgard — Tue, 16 Jun 2026 17:14:48 +0000

I feel the comment about brevity. One of the things that annoys me the most using LLMs is I'll ask it a simple question and get like 2000 words back, or I give it a gentle correction and somehow it takes my one sentence clarification and expands it into 12 paragraphs where it agrees with me and summarizes the thing I just told it in agonizing detail. This is just a thing where I don't think LLMs are good at human context. A person can tell if my question needs a detailed response or a simple one, an LLM generally can't.

New comment by overgard in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

overgard — Tue, 16 Jun 2026 16:41:44 +0000

Curious what model you're using that works well on a 16GB card? I very much want to use my 5080 for inference, but everything I've tried so far has either just not been good enough or painfully slow.

New comment by overgard in "I Love the Computer"

overgard — Tue, 16 Jun 2026 01:59:55 +0000

> AI is better at this than you. You just won't admit it. And it's going to get 10,000x better than you in just a short while.

It's just not though. Plagiarizing some shit it stole off github does not make it intelligent.

Edit: just because it's amusing, here's something I'm literally running into right now with Opus 4.8 on "Max" settings being dumb. I asked it to add some C++ code to an existing C++ project for Unreal Engine. It did half the work, then balked, because "it doesn't have a way to compile C++". I just had to tell it "yes, actually, you just need to run the fricking extremely-standard-already-generated-build-script." If I had a novel build system it'd be one thing, but it already knows I'm working on an Unreal Engine project and that the build is completely standard and it still couldn't piece together that it could just run the compiler.

I would not employ a C++ developer that could not figure out how to invoke the compiler.

New comment by overgard in "I Love the Computer"

overgard — Tue, 16 Jun 2026 01:51:45 +0000

I think a better analogy should be: should you take a job writing a Star Wars movie if you're not a fan of Star Wars? (or sub in: Star Trek, or Dr. Who, or any other recent nerd property that's been kind of ruined by people who don't understand the source material screwing it up by thinking they know better than the fans). In that context at least, the proof is in the pudding: Star Wars is a mess because it's been written by people that mostly don't care about Star Wars, same with Star Trek. I think coding is kind of like that. People that don't care are going to make a mediocre-to-bad product.

New comment by overgard in "I Love the Computer"

overgard — Tue, 16 Jun 2026 01:45:19 +0000

You know what? Gate keeping can be a good thing. Some gates exist for a good reason. I think "are you actually interested in doing this?" is a very reasonable gate. I'm not saying all coding needs to be a passion project, obviously, but I think it's very hard to be good at a thing if you don't care about it. Software continues to get worse and worse, and software developers know less and less about the machines they're working with, and I don't think those things are unrelated.

New comment by overgard in "I Love the Computer"

overgard — Mon, 15 Jun 2026 22:08:14 +0000

Isn't coding solved and we should all be out of a job by now according to Dario? Or what about AI 2027 -- we're only 6 months away! Time to build a bunker!! LLMs themselves aren't snake oil, they're just a useful technology, but all the marketing around them is FUD mixed with hype mixed with the most irritating people on the planet (the ones that aren't bots at least).

New comment by overgard in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

overgard — Mon, 15 Jun 2026 20:01:29 +0000

In my personal experience, I wouldn't bother with 16GB cards for coding -- the useful models are _slightly_ too large to work at any reasonable speed

New comment by overgard in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

overgard — Mon, 15 Jun 2026 19:57:59 +0000

Having a second card doesn't really work well for gaming.

New comment by overgard in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

overgard — Mon, 15 Jun 2026 19:56:55 +0000

Assuming the $100/m claude subscription is still around in three years.

New comment by overgard in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

overgard — Mon, 15 Jun 2026 19:55:17 +0000

I haven't yet, but I just bought a 128GB M5 Max 40 core which I'm hoping can do it (if not, it's a good laptop regardless, I actually need that amount of RAM for non-LLM stuff)

New comment by overgard in "Iroh 1.0"

overgard — Mon, 15 Jun 2026 18:14:09 +0000

This sounds useful, but isn't this the problem that ipv6 is supposed to solve with 128bit addresses? (I'm not really familiar with why IPv6 never really seemed to take off -- does NAT block incoming IPv6 traffic? (I guess that's the other thing -- even though my devices all seem to have IPv6 addresses I can't recall ever using them))

New comment by overgard in "What happened to nerds?"

overgard — Mon, 15 Jun 2026 14:24:48 +0000

I don't think being nerdy makes someone virtuous, that's true. However, I think SV of 20-40 years ago had a distinct culture (best symbolized by Woz) that's basically been lost to the MBA-types and the hustle culture bros. Sometimes they wear it around like a creepy skin suit but it comes across as deeply inauthentic.

New comment by overgard in "Ask HN: What are you working on? (June 2026)"

overgard — Mon, 15 Jun 2026 03:57:21 +0000

The real problem during the pandemic is most of the stimulus money went to the already wealthy. Higher cost of living is because we keep printing money, that money goes to banks, and inflation ends up being an implicit tax on the poor who aren't invested in the markets.

What I can definitely tell you is that the people that currently can't even afford basic things are somehow causing higher cost of living. The economy is not jacked up right now because we gave people laughably small amounts like a thousand dollars. The real problem is people like Sergei Brin spending 57 million dollars to fight a one-time 5% tax.

New comment by overgard in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

overgard — Mon, 15 Jun 2026 00:47:38 +0000

Yes, heavily.

New comment by overgard in "How to earn a billion dollars"

overgard — Sun, 14 Jun 2026 23:46:09 +0000

Here's a fun exercise, go look at the current tech billionaires and tell me which guys (it's basically all guys) you're super stoked about having a billion dollars because they're using it to do so much good for the world?

Because as far as I can tell they're just creating a massive surveilance state while engaging in naked class warfare. Whether you can "earn" a billion dollars, I'm not convinced those people should have a billion dollars.

New comment by overgard in "Not everyone is using AI for everything"

overgard — Sun, 14 Jun 2026 23:35:46 +0000

There's probably a lot more PHP code to regurgitate, while I'm guessing a lot of Swift apps are proprietary. That's the problem.. they aren't coding from principles, they're pattern matching their training data

New comment by overgard in "Show HN: I built 80 mini-games using Fable before it was shut down"

overgard — Sun, 14 Jun 2026 00:24:55 +0000

Who are these imaginary people going to websites and just being like "I'm not spending $10 on this!! I'll just clone it with $500 worth of tokens!"

New comment by overgard in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

overgard — Sat, 13 Jun 2026 06:25:25 +0000

Well, they're selling a non-essential commodity in a highly competitive industry with about zero lock in or customer loyalty, using the standard VC playbook of "capture the market then worry about costs". I don't see space for large margins in there, and if there are margins they're probably recent because of the IPO.

New comment by overgard in ""Don't You Just Upload It to ChatGPT?""

overgard — Sat, 13 Jun 2026 05:28:24 +0000

From what I've heard, personas give a greater chance that the LLM will answer confidently.. and also a greater chance it'll hallucinate something when the data is sparse. Supposedly "grounding" the personas on real documents/web searches is the best approach. Anecdotal though.