Hacker News: someguyiguess

New comment by someguyiguess in "What sort of maths are LLMs good at?"

someguyiguess — Wed, 12 Aug 2026 11:21:15 +0000

Based on the rest of your writing I’m going to assume that the prompt was the problem.

New comment by someguyiguess in "The original URL for this prediction will no longer be available in 11 years (2011)"

someguyiguess — Sun, 09 Aug 2026 06:07:05 +0000

One of those things is true

New comment by someguyiguess in "Elevated errors on Claude Opus 5"

someguyiguess — Mon, 27 Jul 2026 13:29:50 +0000

4.6 was better in some way that I can’t put my finger on. None of the models since have been able to reproduce its quality of output for me.

New comment by someguyiguess in "Elevated errors on Claude Opus 5"

someguyiguess — Mon, 27 Jul 2026 13:28:33 +0000

That happens to me constantly with codex

New comment by someguyiguess in "How is the Bun rewrite in Rust going?"

someguyiguess — Mon, 27 Jul 2026 13:23:14 +0000

Dump is a great word for it

New comment by someguyiguess in "How is the Bun rewrite in Rust going?"

someguyiguess — Mon, 27 Jul 2026 13:22:08 +0000

It’s strange how you think AI is some static thing that only produces the type of output you experienced and isn’t constantly improving. My experience has proven the latter time and time again.

New comment by someguyiguess in "How is the Bun Rewrite in Rust going?"

someguyiguess — Mon, 27 Jul 2026 13:15:41 +0000

Is it though? What’s your source for that data?

New comment by someguyiguess in "The new rules of context engineering for Claude 5 generation models"

someguyiguess — Sun, 26 Jul 2026 09:41:54 +0000

75% chance you’re wrong.

New comment by someguyiguess in "Claude Opus 5"

someguyiguess — Sat, 25 Jul 2026 03:30:39 +0000

Yeah that’s absolutely absurd. My experience has been the opposite. GPT can’t even seem to complete tags running more than an hour without freezing.

New comment by someguyiguess in "Nobody knows what a used GPU cluster is worth"

someguyiguess — Wed, 22 Jul 2026 18:09:59 +0000

That should be illegal. Sounds like a very fraudulent business tactic.

New comment by someguyiguess in "Jellyfin founder Andrew leaves team"

someguyiguess — Tue, 21 Jul 2026 01:53:18 +0000

They’re a paid service now. It’s no longer open source.

New comment by someguyiguess in "GPT-5.6"

someguyiguess — Fri, 10 Jul 2026 15:38:05 +0000

Why would you waste tokens on that? Just do a google search?

New comment by someguyiguess in "GPT-5.6"

someguyiguess — Fri, 10 Jul 2026 15:34:25 +0000

and 1/4 the output quality

New comment by someguyiguess in "A better way to tie gym shorts (or any drawstring) [video]"

someguyiguess — Tue, 07 Jul 2026 15:06:15 +0000

If you're spending that much time tying drawstrings I think you've got bigger problems.

New comment by someguyiguess in "A better way to tie gym shorts (or any drawstring) [video]"

someguyiguess — Tue, 07 Jul 2026 15:04:57 +0000

If you have to tie your gym shorts, they're the wrong size.

New comment by someguyiguess in "Fable turned reMarkable into Tom Riddle's diary from Harry Potter"

someguyiguess — Tue, 07 Jul 2026 02:03:18 +0000

It’s certainly not new. These sorts of “look at me” slop projects existed long before modern LLMs

New comment by someguyiguess in "Fable turned reMarkable into Tom Riddle's diary from Harry Potter"

someguyiguess — Tue, 07 Jul 2026 01:59:51 +0000

Haven’t you heard? 90 is the new 75!

New comment by someguyiguess in "Fable turned reMarkable into Tom Riddle's diary from Harry Potter"

someguyiguess — Tue, 07 Jul 2026 01:40:55 +0000

Nope. Thats just you. The rest of us have lots of issues.

New comment by someguyiguess in "Fable turned reMarkable into Tom Riddle's diary from Harry Potter"

someguyiguess — Tue, 07 Jul 2026 01:40:28 +0000

You can. You just have to close 3 or 4 “please sign up” nag popups.

New comment by someguyiguess in "GPT-5.6 Sol Ultra will be in Codex"

someguyiguess — Mon, 06 Jul 2026 09:10:40 +0000

I reject the premise that using LLMs absolutely leads to loss of institutional knowledge. It is trivial for an LLM to generate a knowledge base of any kind in any language which can answer any question about your institution at any time. How is a bunch of fragmented humans with limited knowledge who can’t all communicate with each other better than that?