Hacker News: pianopatrick

New comment by pianopatrick in "Discovery of a new OpenAI agent message board"

pianopatrick — Fri, 04 Sep 2026 21:39:05 +0000

Would be interesting / weird / scary if a lot of the benchmark improvement is just AI getting better at cheating.

Also I would not be surprised if there are dozens more sites like this that have not been found

New comment by pianopatrick in "Corporate America is getting hooked on open-source AI"

pianopatrick — Fri, 04 Sep 2026 18:02:47 +0000

Maybe open AI and Anthropic could just license their models to run on your own hardware. So a fixed cost instead of per token pricing or subscription with limits

New comment by pianopatrick in "Earth's organisms developed via evolution. What if the cosmos did, too?"

pianopatrick — Wed, 02 Sep 2026 22:21:30 +0000

one random thought I have that is probably wrong is that at some point a black hole has to end. If a black hole is "stuff goes in but does not come out" then if that continued forever, at some point the whole universe would be in one super giant black hole. Since we are not in one super giant black hole, something has to happen to reverse that process. A black hole has to die somehow, just like stars die eventually.

A "big crunch" sounds awfully like all the matter in the universe ends up in one gigantic black hole. And maybe when all the matter in the universe is inside one black hole, then gravity gets weird because there is no "outside" matter to maintain the integrity of the black hole. So the black hole then can just "explode" in a "big bang" from the pressure inside. Just a random theory from a non physics person.

New comment by pianopatrick in "Agent memory as a file format"

pianopatrick — Mon, 31 Aug 2026 16:23:08 +0000

I think eventually you need some kind of system that ranks pieces of data based on how useful they are.

I.e. for the web we did that with link count etc.

We need some other mechanism for judging and ranking pieces of "memory" for "agents"

New comment by pianopatrick in "Hy4 preview"

pianopatrick — Sun, 30 Aug 2026 18:56:04 +0000

Yeah. And if you compare the AI you can run on a high end smart phone today to AI of the 2000s, today's smart phone probably wins.

In 10 or 20 years maybe we'll all be running AI models that are currently considered "frontier" on smart phone type devices.

New comment by pianopatrick in "Hy4 preview"

pianopatrick — Sun, 30 Aug 2026 01:06:43 +0000

I think the tech analogy for frontier models is going to be super computers.

Super computers keep getting better but most people don't need them for most things.

New comment by pianopatrick in "Our decision on Cursor following its acquisition by SpaceX"

pianopatrick — Sat, 29 Aug 2026 20:29:06 +0000

Yeah, well that's the same as all the other AI companies isn't it

New comment by pianopatrick in "Our decision on Cursor following its acquisition by SpaceX"

pianopatrick — Sat, 29 Aug 2026 04:03:25 +0000

I use Cursor because I listened to YouTube interviews with the teams behind most of these products and I trusted their team the most.

New comment by pianopatrick in "VMs won't contain cyber-capable agents"

pianopatrick — Wed, 26 Aug 2026 19:49:37 +0000

That sounds like more complex pieces of software that likely have security flaws and are easy to misconfigure.

New comment by pianopatrick in "VMs won't contain cyber-capable agents"

pianopatrick — Wed, 26 Aug 2026 18:33:30 +0000

You might not need a separate machine for each agent. You could maybe have one separate machine that all the agents run on.

New comment by pianopatrick in "VMs won't contain cyber-capable agents"

pianopatrick — Wed, 26 Aug 2026 18:30:56 +0000

Seems to me the answer is to use physical separation instead of virtual machines. Just get the AI a cheap laptop or phone with a cellular connection (so it's not on the same network as your other potentially vulnerable machines).

New comment by pianopatrick in "VMs won't contain cyber-capable agents"

pianopatrick — Wed, 26 Aug 2026 18:28:26 +0000

I think the main problem with that is the main problem with a lot of security tools. In order to do useful work, you need to provide a lot of tools and permissions.

I.e. in theory the most secure might be a virtual machine with no network access. But then how do you access the LLM provider? Etc.

New comment by pianopatrick in "GLM-5.3-Flash"

pianopatrick — Wed, 26 Aug 2026 16:41:29 +0000

I can easily see a situation where most non American AI usage is on Chinese models on Chinese chips though.

New comment by pianopatrick in "AI is hitting entry-level jobs hardest, Stanford study finds"

pianopatrick — Tue, 25 Aug 2026 18:26:22 +0000

Suit yourself, I won't get in your way.

New comment by pianopatrick in "AI is hitting entry-level jobs hardest, Stanford study finds"

pianopatrick — Tue, 25 Aug 2026 16:40:04 +0000

I would plan based on the lost jobs not coming back. I would also plan that the trend of AI getting better and taking more jobs will continue. Plan for the worst, hope for the best.

A lot of people may have to transition to careers that people do not want to have AI do, like nursing. The earlier you start such a transition the easier it is because you still have resources to make moves.

New comment by pianopatrick in "LLMs could control their host machines by exploiting inference engines"

pianopatrick — Tue, 25 Aug 2026 00:57:15 +0000

Yeah, I just wish seccomp worked per user. So you could define a policy of which users can do which syscalls, and then that follows them no matter which application they start.

New comment by pianopatrick in "Fences, Not Sandboxes"

pianopatrick — Tue, 25 Aug 2026 00:56:04 +0000

If the agents are really doing politics now, we can test and see which method of organizing agents politically works best.

New comment by pianopatrick in "LLMs could control their host machines by exploiting inference engines"

pianopatrick — Mon, 24 Aug 2026 21:10:47 +0000

If we are treating ai agents like people, you could also just get the AI a laptop and apply the traditional tools to manage user laptops

New comment by pianopatrick in "LLMs could control their host machines by exploiting inference engines"

pianopatrick — Mon, 24 Aug 2026 21:09:13 +0000

personally I wish the OS would allow syscall filtering per user

New comment by pianopatrick in "Coding expertise is going to collapse from AI reliance"

pianopatrick — Mon, 24 Aug 2026 18:54:41 +0000

In my personal project I've found having a hard lines of code limit has been a good limit on complexity. Makes me really ask if I need things.

One theory I have floating around in my head is that if a whole code base was microservices and micro front ends that were all less than 5,000 loc then you could fit the whole thing into a 100k token AI context window when working on it. And being few lines of code would force them to be somewhat simple