Hacker News Reader

You have to accept that people are different than you.

I see long massive pro apple threads. I don't get it at all. As in; literally don't understand what Apple is good for. But I have friends, family irl who love apple so I know the sentiment exists. I therefore accept that many HN users are similar.

Many people really truly are happy to see another model drop and are excited about progress etc etc. Surely you've met such ppl in real life. Well, they're here too (I'm one of them fwiw)

FergusArgyllon GPT-6 Sol and Luna

Oh, I'm happy I'm not the only one. Astra was feasting on tokens!

FergusArgyllon Claude Opus 5.5

Well, they're actually older so it makes sense that their model versions should be ahead

FergusArgyllon Claude Opus 5.5

It has read the internet. That doesn't mean it was literally RL'ed for this

FergusArgyllon Grok 4.7

https://www.dwarkesh.com/p/pretraining-progress-is-mostly-da...

An amateur but worth reading nontheless

FergusArgyllon US Military had close call after using AI for hallucinated intelligence report

What is the purpose of an army?

FergusArgyllon OpenJev

Yeah, it outputs probabilities over the options you give it. It's basically a generalizable classifier.

FergusArgyllon Fable 5.1 Solves the Cyphral Distich, a 370-year-old cipher

Theoretically, with a long enough codeword to cipher ratio, vigenere is impossible to break (I think) (but sort of useless)

And yes, I've taught 8 year olds how to crack Caesar ciphers...

FergusArgyllon Fable 5.1 Solves the Cyphral Distich, a 370-year-old cipher

It's good at poking holes at my galaxy brained newfangled ideas for ciphers too. I thought I had something good, pasted the ciphertext and got a "it was embarrassingly simple..."

FergusArgyllon So you want to use OpenRouter?

It's not always deceptive. Bugs in inference can be very subtle (well, bugs everywhere can be subtle).

The providers should be benchmarking their offerings daily

FergusArgyllon A misalignment of AI in mathematics

The cheapest kind of journalism. Acme corp does x, Scroll twitter, find malcontents, write an article "People are upset about x".

I'm outraged at mathematicians. There, write the article.

FergusArgyllon Show HN: Bodily Oddities

Huh, call of the void is common and it has a name!

Great site

FergusArgyllon Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra

There are multiple kinds of distillation

https://arxiv.org/abs/2106.03310

https://arxiv.org/abs/2207.12106

FergusArgyllon Stockfish 19

Way way shallower search. A human doesn't consider more than a few candidate moves. But they (we?) build up much better intuitions and heuristics

FergusArgyllon Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra

Distilling you don't have the actual model weights of the teacher. All you have are the teachers answers to a lot of questions. You then teach your own smaller model to answer more similarly to the big teacher model.

Fine tuning you have the actual model weights of the original model, you then train that model to answer in a different (or better) way.

FergusArgyllon Discovery of a new OpenAI agent message board

When I studied cybersecurity I always had this nagging feeling "But I can never get past something real" Turns out I can bypass the defenses of a trillion dollar tech company!

FergusArgyllon Formalizing Fermat's Last Theorem

Ooh I never realized FLT and Code book were the same author. Yes, both great!

FergusArgyllon Gemini 3.8 Flash and 3.8 Flash Cyber

If that's not art, I don't know what is.

https://scosman.github.io/pelicans_riding_bicycles/

FergusArgyllon Google Has Removed MV2 Extensions from the Chrome Web Store, Including UBO

eww?

:r !curl?

FergusArgyllon OpenClaw 2.0, Accidentally

This was useful for the 1-2 months where models were good enough to use all their newfound tools but not good enough to vibe code them themselves.

Now, just install codex (or claude code or whatever) on a vps, tell it to make a way to email it, then email it to make a way to text it etc etc. takes a few minutes and when you need another bridge just have codex make it.

FergusArgyllon Creepy Crawlies

Woah, went from > 1 min on git.ffmpeg (difficulty 6) to instant. Thanks!

Edit: after some more analysis; Average with extension disabled is 90 seconds, with extension enabled it's .3 seconds

FergusArgyllon Nvidia projects $673B in sales as AI demand widens

Hermes is close

FergusArgyllon Ask HN: Let's all sell our AI stocks, short Nvidia and pop the AI bubble

At least AI is not depressed all the time and trying to make everyone lose their money

FergusArgyllon GLM-5.3: Frontier coding with emergent cyber capabilities

I think there's still a ton of secret sauce needed for serving them economically

FergusArgyllon GLM-5.3: Frontier coding with emergent cyber capabilities

More RLVR. Give it verifiable problems, if it doesn't find a solution move on, if it does, use that as a reward signal.

FergusArgyllon Launch HN: Bullet (YC S26) – A Faster Coding Agent

They shoot through walls!

FergusArgyllon "AI" will never become conscious

Would be nice if we can take the bat's sonar encoder, train a projector and fuse it into our NN

FergusArgyllon Ask HN: What's Next for LLMs?

Yeah, so it's a spectrum. In general if you want to feel the power of the latest and greatest, you should just not look at the output, give it full access (not on your computer - a server, or old machine you don't use is best) and give it a big audacious goal and let it work.

That might not be ok for your work / hobby / whatever, but if you wanna see the differences between models that's probably the best way.

FergusArgyllon Show HN: Alfa. Killing AI hallucinations with resonance

> If we represent the operation of the Random Number Generator (RNG) as a time-series graph, the system can be applied to decrypt the bitcoin seed-phrase.

Big if true

FergusArgyllon Ask HN: What's Next for LLMs?

If you can't find a difference between Haiku & Opus, something's going wrong