Or just.. download the books, and slap a search engine on top of them. It'll be like 10x smaller than the model. "Survival LLM" is absolutely a non-relevant use case.
19,483 karma · joined October 14, 2012
There's some genuinely awesome stuff out there, but it's (paradoxically) much rarer imo than it was a decade ago. HN and Twitter is a place where it occasionally bubbles up.
It's because most content worth seeing these days is behind a walled garden: Twitter, reddit, TikTok, Instagram, Snapchat, YouTube, and even news articles these days are often paywalled. It's just a fundamentally different internet than it was in 2009. The corporations have definitively won.
One reason I personally like AI is that it at least puts the end-user on the same level (at least automation-, scraping-, and compute-wise) as big companies.
[1] https://www.npr.org/2026/05/13/nx-s1-5819609/federal-governm...
In fact, I would counter and say that Hammurabi's Code is a quintessential example.
This is absolutely not true, and I would strongly suggest taking an ethics course. One of the most fascinating realities of the human condition is that it does seem that moral imperatives tend to be constant, the most obvious of which is the Golden Rule (which is at least 3000 years old), but there are many (the Ten Commandments or Hammurabi's Code are good jumping off points).
In fact, the proof is such a mess, nobody really understands how it works. Which is becoming commomplace with these permutation machines where we can come up with something that passes some tests (e.g., a proof checker), but we have no idea wtf we did to get there, which doesn't really seem to do much to push human understanding forward.
It seems like such a dumb hill to die on, too. OpenAI literally stole the entire Internet (including copyrighted works), and we're arguing about a silly math puzzle.
The license doesn't really make much sense to me, because if you're already at $10m revenue, you can probably train your own (better, more tailor-suited) model, or at the very least raise money to do so.
Of the ~decade since I've seen these types of apps, I've never used one and, frankly, I never ever craved for "just sharing one window" or "playing in multiplayer" as @jsherwani put it. Which is kind of odd, because COVID came and went (peak remote/screen sharing days), and I had totally forgotten these kinds of products even existed. While this is (still) super neat technically, I just don't really see a concrete use case other than "lemme show you something super cool."
Has anyone actually used it? Where would this kind of tool be better than just screen sharing over Google Meet or Discord or Slack or whatever?
You're right, I'm sure the people that are publishing these studies have never heard of "correlation is not causation."
Millenial cynicism is so boring.
People have always done this tbh, I've probably built like a dozen+ games at game jams that even I barely remember.
> It appears that GLM-5.2 is able to propagate Llama3.1-8B well, which suggests that future small models of the same capability level could also complete the network enumeration, payload delivery, and inference standup steps of the replication process.
"Able to propagate" is (the paragraph above): "accessing them via a reverse shell, sending over the viral payload (weights and harness), and setting up inference." The thing isn't able to do anything, it's running shell commands it was told to run. "Oh my God..." indeed. What are we even doing here? This blog post is embarassing: "Look, you can run stuff on compromised non-firewalled recklessly open systems!"
I have to constantly make sure it’s building things in a way that will fit neatly together, in a way doesn’t over (or under) -generalize. In a way that doesn’t waste time and energy on idiotic tests that aren’t really testing anything at all.
LLMs are powerful autocomplete with vast and broad memory, they do not think. And thinking is just a part of the human condition.
Kind of sad seeing so many “tech people” (including here on HN) falling for it hook, line, and sinker.
If I seriously needed a classifier, I would just train my own and it would run on an iPhone. Any CTO worth their salt would suggest the same, because it's not even remotely comparable to training a large language model (w.r.t. compute or training data required).
Invisible Cities is one of my favorite books, and I've re-read it several times. It's a great "travel" read and always puts me in the mood to explore. It's also a very romantic book, obviously the cities have all named after women, so it evokes conquest in the sexual sense as well. It's also a piece of surrealist/allegorical literature in the same vein as Daumal's Mount Analogue (which also tops my list).
Back to the website. While this is an impressive tech demo (as they all are), it's clear that AI misses the forest for the trees, and here's a tangible example. I randomly clicked on Theodora. Honestly, I didn't even remember which city this was until I re-read the Calvino passage:
> Recurrent invasions racked the city of Theodora in the centuries of its history; no sooner was one enemy routed than another gained strength and threatened the survival of the inhabitants. When the sky was cleared of condors, they had to face the propagation of serpents; the spiders' extermination allowed the flies to multiply into a black swarm; the victory over the termites left the city at the mercy of the woodworms. One by one the species incompatible to the city had to succumb and were extinguished. By dint of ripping away scales and carapaces, tearing off elytra and feathers, the people gave Theodora the exclusive image of human city that still distinguishes it.
So, what do we know about this city? We can make a short sensible list:
1. It's an imperialistic (or at least millitaristic) reflection of man's will to power (as Nietzsche might put it)
2. It's decidedly anti-nature, the natural flora and fauna being its primary antagonist
3. It's been destroyed and rebuilt many times
4. The city is built by literally ripping away bugs and birds; tearing their wings from their bodies like a cruel child might
5. Nature is slowly winning (and maybe it already has)
Calvino, again, on Theodora:
> But first, for many long years, it was uncertain whether or not the final victory would not go to the last species left to fight man's posession of the city: the rats.
The vibe here is an overrun rat's nest of a city. A disgusting bloated corpse; more:
> The city, great cemetery of the animal kingdom, was closed, aseptic, over the final buried corpses with their last fleas and their last germs.
I see an overrun New York. I see a city built on a decaying graveyard. AI sees animals and humans living in a kumbaya harmony, but that is not what Theodora is: it's an eternal conflict between man and nature (and nature won a long time ago). For the record, this isn't even that particularly deep. Many authors and poets write about this eternal human struggle: the taming of the wilds. But Claude just doesn't get it. I hope readers of the book do.
Anyway, I leave you with this. I found an artist, Rian Hotton[2], who put brush to canvas to show us what he thought Theodora looks like[3]. And this... this is more like it.
[1] https://imgur.com/a/hXrKVg3
I’m interested in this because “macro maps” can be hand built in a way that may want to preserve gameplay balance while individual games can still feel broadly unique.
Running a decision model is way easier and much cheaper. Are they really just trying to capitalize on the hype here? It feels like they really have absolutely zero moat.
Claims like this needs to be deeply analyized. Models do have some emergent capabilities[1], and I think there's a lot of evidence to show that semantics is actually learned (Word2Vec), and some math seems like it also might be learned (e.g. modular arithmetic). But it's hard to exactly say where there's some internal mechanism generating a true answer and where we're just getting lucky with some distribution so the answer just seems right.
It searched for "Aliexpress" on Google and clicked the link, hence the Google UTM params.