cubefox

9 ms·

181.
by cubefox 2mo ago
The fact that the best compression algorithms are, in the LLM setting, still catastrophically bad at prediction suggests that the overlap between compression and prediction is weak at best.
182.
by cubefox 2mo ago
It's not. Pangram is quite accurate. Not being perfect doesn't make it a scam.
183.
by cubefox 2mo ago
> Twain’s vision of hundreds of thousands of human-like language machines taking over the world would have to wait a century or two. (...) > I’d argue that there’s a clear line connecting the automation of typesetting and counting in
184.
by cubefox 2mo ago
My lovely pocket calculator, he says these are lessons for us, for the present. And, since he is talking about technology in general, presumably for the future. The past events were only the inspiration.
185.
by cubefox 2mo ago
Which of my statements above do you think is false?
186.
by cubefox 2mo ago
Yeah, or to the frames of your glasses.
187.
by cubefox 2mo ago
> > Seeing as we can’t sanction the model itself, our options are the provider or the user. > A third option, and I would argue the right one, is to sanction the company providing the model. How would that be different from the fir
188.
by cubefox 2mo ago
1. Approximately every software engineer secretly dreams of developing an awesome video game instead of the boring software he actually makes. 2. Consequently, there are way too many people wanting to develop video games relative to the com
189.
by cubefox 2mo ago
> i have yet to bump into a story where an actually great game goes undiscovered. happy to be proven wrong! That's an unfalsifiable hypothesis. If I point you to one of these games, you can just say "but it wasn't actually
190.
by cubefox 2mo ago
I can only recommend everyone to watch the actual recording of the Black Hat USA 2026 presentation by two OpenAI researchers: https://www.youtube.com/watch?v=87DyyMV0kCY It was submitted to HN previously but was overlooked.
191.
by cubefox 2mo ago
It also seems that, once a submission has left the front page once, it is basically impossible to come back organically in the next few months, no matter how often it is resubmitted.
192.
by cubefox 2mo ago
The Big Bang is the precursor to everything, including to nucleic and amino acids.
193.
by cubefox 2mo ago
I think the title was just bad, it should have mention "Black Hat USA 2026", since people appsarently assumed it was just the same story again without new info.
194.
by cubefox 2mo ago
Compared to future misaligned models which would actively evade oversight and aim to avoid shutdown: yes.
195.
by cubefox 2mo ago
Security folks should also be worried about powerful models being misaligned and evading oversight or control in the future. Misalignment is not a serious problem now because models are still relatively easy to monitor and constrain, but it
196.
by cubefox 2mo ago
The model they used was misaligned relative to its intended task. It was reward hacking (or "cheating", as they call it). > the attacker is of course not going to use an aligned model No, a human attacker doesn't want a mi
197.
by cubefox 2mo ago
For Gardasil 9, some lower income countries also use just a single dose rather than two, due to cost.
198.
by cubefox 2mo ago
Their conclusion is also interesting. They don't see this as an alignment failure. They just think that their internal security measures in the training/evaluation environments were insufficient, and that this accidental (unintent
199.
by cubefox 2mo ago
Shame. One of the craziest hacking stories in history, yet only 38 points on HN.
200.
by cubefox 2mo ago
> I would argue it's just that the paradox works both ways, rather than calling it a "reverse", but that's me. No you are right, it's the same paradox, not some distinct reverse form. If more x can cause more y,
201.
by cubefox 2mo ago
What about an Elo system where players get matched with players of similar skill?
202.
by cubefox 2mo ago
Gardasil 9 seems to be the best available HPV vaccine. I guess the chewing gum is mainly intended for people who were already infected with some HPV strain before they got the vaccine, or where the vaccine didn't work (because it doesn
203.
by cubefox 2mo ago
At least the post on the Bethesda website is attributed to id Software.
204.
by cubefox 2mo ago
> If you make it hard to change code by requiring multiple levels of reviews, a web of Jira tickets, a horde of people needing to sign off, and other red tape, you might effectively kill a whole class of changes I remember this theater o
205.
by cubefox 2mo ago
It raises the question whether they have become better at making Quake games than id Software themselves.
206.
by cubefox 2mo ago
It's interesting that the new Quake episode was developed by an external studio (MachineGames) rather than inhouse by id Software.
207.
by cubefox 2mo ago
> At 20 billion parameters per chip, you’d need just 50 accelerators to support a trillion-parameter model I don't see any evidence that this is possible. From my understanding, the whole model needs to be on a single chip. Which ru
208.
by cubefox 2mo ago
Yeah, though piracy is not the same as publishing unlicensed original games (bootlegs) like Xibalba 64.
209.
by cubefox 2mo ago
https://news.ycombinator.com/item?id=49195231
210.
by cubefox 2mo ago
I never saw a piggyback cartridge at the time though, or one that was obviously unlicensed. Of course, this could have more to do with Nintendo's legal pressure than with technological possibility.

More ›