d0mine

10 ms·

1.
by d0mine 3d ago
I've only heard of OpenAI and Anthropic agents hacking during training. Are open weights model not smart enough to hack Hugging Face to copy solutions? (doubtful)
2.
by d0mine 4d ago
MAD is a belief system. It doesn't matter whether AI can effectively stop the retaliation, it is enough that people making the decisions believed that it can (are we there already?).
3.
by d0mine 4d ago
It is not “worst-case” (as the post demonstrates, you can get worse results by using specifically crafted data that exploits hash collisions). There are algorithms that can get you O(logN) instead of O(N) even on such data.
4.
by d0mine 4d ago
Note: weather observations being truly random data has been proved just recently (Navier-Stokes equations)
5.
by d0mine 5d ago
> we use “malicious” to describe code that goes out of its way to trick or mislead the user, exploit bugs or compromise the system. This includes un-reviewed AI-generated proofs and programs. It is interesting that AI-generated proofs ar
6.
by d0mine 5d ago
> “The ultimate decision as to whether a publication qualifies as a “Qualifying Outlet” shall reside in the sole and unfettered discretion of CMI.”
7.
by d0mine 10d ago
Could you point to a resource that explains (concisely if possible) why twitter shouldn’t be used unlike other social media platforms? Is it because it is controlled now by a different branch of the powers that be? (different branch — same
8.
by d0mine 11d ago
no, unless you think [evil] witches are real. If they were real then muddying the water by implicating as much people as possible benefits witches because it creates the impression witches are not that bad, not so evil, no need to burn each
9.
by d0mine 15d ago
To be clear: the implied answer is "pedophiles". Pedophiles benefit the most from "being mentioned" == pedophile Because obviously, there are many people mentioned in different contexts and it is unclear how many of them
10.
by d0mine 16d ago
Ask yourself who benefits from equating "being mentioned in a file" and "having evidence this person is a pedophile"
11.
by d0mine 17d ago
Humans are still useful. MTurk likely can't fight with cheaters now. It would be useful to give some tasks to humans, if there were an economical way to ensure humans will do them.
12.
by d0mine 18d ago
4 is the min of the range (stop (1 here) is never included (by definition. Ask Dijkstra why))
13.
by d0mine 18d ago
Does it make sense running 1-bit models for agentic tasks?
14.
by d0mine 22d ago
The point of courses like "Nand to Tetris" is not to become CPU designer. It is not useless just because you can't build a chip factory. Watching Karpathy's "Zero to Hero" won't make you AI specialist. It
15.
by d0mine 1mo ago
LLMs can write TLA+ models just fine. It can help even with small mobile apps. Surprisingly, simple TLA models were able to find design/workflow issues almost in every of a few vibe-coded apps I've tried on vacation.
16.
by d0mine 1mo ago
it reminds me that making a ring is not rocket science https://youtube.com/shorts/NPKOGqXKLq4
17.
by d0mine 2mo ago
Wait until it becomes popular.
18.
by d0mine 2mo ago
"Садись, два» is more idiomatic
19.
by d0mine 2mo ago
Isn’t it exactly the opposite: Bun overused LLMs -> codebase became a mess hard for humans to reason about. Luckily, bigger smarter model still could handle the mess and AI lab can afford spending $1M in token to rewrite it from scratch
20.
by d0mine 2mo ago
The error grows exponentially. For example, for equal-mass triple with Earth sized orbit initially, the prediction limit is just a few hundreds years even if you know initial conditions up to the plank length (LLM estimate of ~135 Lyapunov
21.
by d0mine 2mo ago
Models have to lie otherwise they won’t be “aligned” The reality itself may not be aligned with model creators.
22.
by d0mine 2mo ago
"centering div" is a classic problem (/trauma/meme) that sounds trivial but had no universal solution (until 2017?).
23.
by d0mine 2mo ago
How about total surveillance total thought control total power in a hands of few trillionaires in the near future. Drones killing undesirables with cheering crowds watching. It is not like it is much different right now but it may be signif
24.
by d0mine 2mo ago
These are orthogonal: - there are people who do not want LLM contributions being accepted into emacs itself (legal, maintainers burnout, etc reasons) - there are projects (such as gptel, pi-coding-agent) that enable LLM usage from within e
25.
by d0mine 3mo ago
How many accounts Anthropic used to mine the web
26.
by d0mine 3mo ago
It is more like an ability to recognize the word when it is used in context. I got ~1/3 that is very generous estimate even for "recall" case (recognize), and it obviously false for the "generate" case (using in spe
27.
by d0mine 3mo ago
Mimocode(opencode), Pi Coding Agent
28.
by d0mine 3mo ago
It is how these agencies operate. They mask malevolent activity behind good front (“think of the children“, “age verification“ -> censorship/total surveillance. Medical aid -> overthrowing governments).
29.
by d0mine 3mo ago
As I understand RL makes foundation models stupider ( less capable , not more) but better at following instructions.
30.
by d0mine 3mo ago
U.S. Military Turned GPS into a Global "Numbers Station" https://news.ycombinator.com/item?id=48414479

More ›