ozgung

3 ms·

1.
by ozgung 5d ago
This will be the same story for every industry again and again. AI is not good in x=Finance because models were not RL trained heavily on x=Finance capabilities yet . This is only because Big Labs have finance benchmarks lower in their pri
2.
by ozgung 6d ago
LLMs solved this exact problem. Not perfectly (they can sneak in some malicious commands) but good enough for everyone. This is also mostly what non-developers mean by "vibe-coding". They can "tell the computer to do the thin
3.
by ozgung 8d ago
Isn't that the same transformer at the end of the day? It must be faster only because it generates a single token output, just one evaluation of the model. It takes the same input context and has the same O(n^2) attention blocks. It pr
4.
by ozgung 8d ago
They came for software first. It was the industry that was “eating the world”. Mathematicians were also affected. Because math was also “code” thanks to Lean. It was collateral damage. Now they’re coming for lawyers. Law also has “code”. It
5.
by ozgung 9d ago
OpenAI case was actually an exception. Agents had no internet access because they were evaluated for a benchmark. In real life agents have access to virtually everything, most people use them like that. If any of you actually know how to ma
6.
by ozgung 10d ago
Both. Two things can be true at the same time. It’s a classical regulatory capture attempt by the major players, rather than a marketing ploy. They use safety as an excuse. AI agents can generate and run arbitrary code. They are very capabl
7.
by ozgung 12d ago
Source? How do you know they were "prompted to hack to get answers"? How do you guarantee they will always listen to you when you say "do not hack outside systems". They are not classical deterministic programs doing exa
8.
by ozgung 12d ago
The answer is: historically the incumbents define the rules and use safety as an excuse to build their moats by gaining exclusive rights and they end up being cartel/monopolies. It’s a rhetorical question.
9.
by ozgung 13d ago
It's like we give them Asimov's Three Laws of Robotics and robots say "nah".
10.
by ozgung 13d ago
> were intentionally misaligned or had guardrails turned off I think the bigger story is: Guardrails don’t actually work and we can’t align these things.
11.
by ozgung 13d ago
TL;DR: Researchers assumed it would be easier to align these models when they become smarter. In reality, agents are getting increasingly smarter but aligning them also becomes increasingly harder, or even impossible. They are losing their
12.
by ozgung 13d ago
I’m pretty confident this has already happened. Public models are behind their internal models and just above Chinese models. Only thing closing the gap is Chinese models pushing. I’m not American. There is no way I use the same model as US
13.
by ozgung 14d ago
There is a small web service I need with a price tag of $39 one time payment. I’m current doing it myself for the last 1.5 hours, because I thought it would be fun. 2 minutes ago I upgraded to $120 subscription of Codex. Reasonable thing wa
14.
by ozgung 14d ago
Humans DO conspire. Interest groups and politicians DO conspire. Governments DO conspire. Big corporations DO conspire. When you have the mix of everything above, they definitely conspire. “Theory” part is only about finding out WHAT the ac
15.
by ozgung 14d ago
And this is exactly the point of the Statement. The process is more important than the solution itself. Most problems in mathematics don’t have immediate value or applications to the real world. AI’s solutions are like the answers section t
16.
by ozgung 15d ago
They say it’s a violation of TOS in Claude’s case. I’m not sure books can have TOS to protect themselves. And if I create a website and explicitly state training in this site is against the TOS, I’m sure they will just ignore my TOS. Legal
17.
by ozgung 15d ago
More interesting dialogue was between the Claude Executive and a government official just before this one.
18.
by ozgung 15d ago
You’re right. I used quotes when I was really paraphrasing. Here is the actual paragraph from the statement: > I said that if OpenAI released its result in the way proposed I would go public with what happened. The reply was, “Why would
19.
by ozgung 15d ago
Here is a new rumor for you: I and my collaborator who is a leading math professor in this specific area are very close to solving another Millenium Prize problem, Hodge Conjecture. We’re working on this since last year. Already proved some
20.
by ozgung 16d ago
No. 1. Buckmaster contacted OpenAI first. Not the other way. 2. Giving the $1M bounty to a human mathematician for the effort and giving him credit would be excellent PR. They had already burned much more than $1M for the generation. Adding
21.
by ozgung 16d ago
Always has been. Secret History of Silicon Valley by Steve Blank https://youtu.be/ZTC_RxWN_xo
22.
by ozgung 16d ago
Very likely. These mathematicians’ prompts are not like “hey chat, please solve Navier-Stokes for me”. They add real expertise and intuition from the cutting edge of their field.
23.
by ozgung 16d ago
If your rumor is true, what we are witnessing is a giant paradigm shift rather than individual incidents. Mathematicians were the first victims of super-intelligence. Of course it’s not an endless source. They had to burn millions of dollar
24.
by ozgung 16d ago
Happened to me recently, but on Claude. I resubscribed to Claude Code two weeks ago for a side project and updated it. I checked for the setting after these last events and it was turned on. I'm sure I checked them few months ago. Ther
25.
by ozgung 16d ago
11 Pro Max. First battery needs replacement but it's ok on iOS 27. I don't need or want a new phone.
26.
by ozgung 16d ago
This is actually bigger than that. This is an "AI is eating the world" situation. The math community was one of the first to be affected because RLVR makes Math an easier target for AI. We witness AI eating the math community. Sof
27.
by ozgung 16d ago
In the old torrent days I was only watching classical masterpieces and top level world directors. In the streaming era I only watch mediocre straight-to-streaming movies, few popular TV series and too much reality TV that I'm not proud
28.
Ask HN: What is the proper way of preventing AI labs to use your private data
1 points by ozgung 16d ago|1 comments
29.
by ozgung 16d ago
I think Sam Altman already admitted that is a possibility.
30.
by ozgung 17d ago
Correct. But… According to data in the site and paper [0] labor income is estimated $20T for all four scenarios (including no AI) in 2030. Which means capital takes ALL of the new pie coming from AI. Also, I am not an economist but I assume

More ›