HNHacker News
TopNewBestAskShowJobs

chr15m

2,073 karma · joined July 8, 2015

comes with absolutely no warranty

https://mccormick.cx

https://github.com/chr15m

https://youtube.com/mccormix

submissionscomments
chr15m··on Butlerian Jihad
Although LLMs are not deterministic in normal operation (because of the temperature values used) they are still strongly biased in their outputs. Bias in their outputs is literally how they function.

So you get a different type of conformism where people who interact with LLMs a lot are going to be influenced towards the median, as encoded in those token probability look up tables.

The effect is basically the same as what Heidegger and Illich were worried about, and I think we're seeing it happen already everywhere.

A solution(?) is to try to be your true weird authentic self when you act in the world, rather than bending your words and behaviours to the machine and LLM probability shape.

chr15m··on Why is every rationalist blog post way too long?
Gish gallop. Harder to argue with.
chr15m··on A Modern Web App in C
Software developers love complexity. They love to construct huge beautiful towers of the stuff.

Also it's the code version of this old phrase: ("I Had More Time, I Would Have Written a Shorter Letter") https://quoteinvestigator.com/2012/04/28/shorter-letter/

chr15m··on Passkeys are a trap
Changing banks is incredibly annoying but this might be a reason to consider it.

I have also faced this issue with banks and their frustrating authentication issues. Not to mention proprietary app requirements.

chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
The harness is using my own project "runprompt" which you can find here:

https://github.com/chr15m/runprompt

That has a bunch of Python tools for things like adding to TODO list, notes, looking stuff up, etc.

chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
Things I use the private inference machine on my local network for:

- Running queries on bank transaction data to figure out our home budget, savings and spending rates, etc. I don't want this data in the hands of AI companies.

- Joplin note-taking app LLM integration for searching, creating, summarizing notes. This is my private notebook and I don't want it leaking.

- Termux based phone assistant. I can use voice to ask it to add something to the shopping list or my notebook. Much faster than pulling up the app. Especially useful when out walking. Like a low-key home-grown Siri. I'm adding more skills to this as I need them.

- Building a little robot character in our home that people can use to ask weather etc. This is mostly an experiment so far.

- Various other one-off use-cases and experiments. Recently I've been testing out Clef which is like a self-hosted Jev that can read images. Response time of 300ms it's wild! Haven't found a use for it yet.

chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
Yep, home made. It's not online. I will publish it at some point, but it's a very simple web app you could put together pretty easily.

Here's what it looks like: https://mccormick.cx/gfx/blogref/priority-dashboard.png

The projects are stored in a YAML file with entries for name, source, and desired frequency.

Then the script just ranks them by how overdue they are relative to the desired frequency.

The source field can point to an RSS feed (for YouTube and blog), a git working folder, or any other type of folder. It basically checks if there's anything new depending on the source type and that's how it knows something has been worked on.

For house maintenance tasks you could simply tap to mark something as "I did this today" and have it save to localStorage or whatever.

chr15m··on An open-source, clean-room reimplementation of Microsoft Word in pure Rust
> extremely buggy alpha state

I'm curious about this part. As a user you're finding them extremely buggy - what kind of bugs are you seeing?

It sounds like you're expecting the bugs to be ironed out in six months. I guess I'm wondering if the same process that led to buggy software in the first place can actually be used to get rid of the bugs without creating more.

chr15m··on Passkeys are a trap
An uncompromising minority has surprising power. See Nassim Taleb's idea of assymetric minority rule.

Passkeys are only one kind of 2FA.

Yes, banks are particularly bad at this because their threat model is strong.

My bank required installing an authenticator, but apparently recently they've allowed SMS based 2FA. It seems users complained/resisted and they had to accommodate them. Maybe you should give feedback to your bank. Be the uncompromising minority.

I think 2FA is fine and a good idea, but I agree with the author that passkeys seem fraught as designed.

chr15m··on Passkeys are a trap
If you don't like passkeys the solution is very simple. Resist using them. As long as a significant minority resist and refuse, they will have to keep other types of auth around.
chr15m··on The Lightbulb Computer
Just want to say, what an incredible vision. Congratulations and thanks for sharing!
chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
I guess you mean "openai-weil". Thanks, that's fascinating! I had no idea about the competing conventions thing.
chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
I didn't mean to come across as saying people are perfect. I 100% agree that human beings are often catastrophically wrong, even when they have every incentive to be correct.
chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
Ah ok, so the particular subset of kernel soundness bugs that would affect con leche, understood. Thanks.
chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
> If AI advances more and becomes smarter it can be used on both sides.

Yes, and that's true even if it doesn't advance or become smarter. It's happening already that people are using AI on "both sides" both adversarially and defensively and that is improving software, and hopefully mathematics as well.

> I am not sure about the Navier-Stokes proof but in general it sounds highly unlikely that many well-known theorems turn out false.

I'm talking about the LLM Navier-Stokes proof specifically because it's very long and complicated and an LLM is not the same as a human mathematician. A human mathematician is trying to use the tools in a truth-seeking way, whereas LLMs are trying to satisfy the goal of presenting a proof that convinces people it is correct. Maybe naively, it seems to me the failure modes of proof assistants would be amplified under adversarial conditions (i.e. driven by LLMs).

Completely agree that there seem to be many ways that relying on proof assistant software can go wrong.

chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
I am conflating (probably ignorantly) the idea in software development of bugs in recursive code, with the potential for bugs when using Lean to check Lean’s checker.
chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
Are you saying it's very unlikely that further lean kernel bugs will be found?
chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
> without having the necessary background... it is difficult for people to engage with you since we have no idea whether what we say is understandable by you.

That must be frustrating. Thank you for trying and engaging with me anyway.

Those links were very informative, thanks for sharing!

It seems like there are a lot of things to worry about (even more than I realized after reading the main article) when using a proof assistant to try and figure out what is true. As they point out in the material you linked to, this is particularly fraught if you can't assume the prover's honesty (per your step 1), as is the case when it comes to LLMs.

From what you've said and linked to it seems like you agree with the thrust of my original comment, or are you just saying "mathematicians already realize all of this you idiot," in which case I accept the charge!

I don't think endorsing scepticism and epistemic humility is a glib statement. These are fundamental mental tools when trying to figure out what is true.

chr15m··on Programming Isn't Special
Art, or craft? Or both?
chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
It seems like very important work and was discussed in the article with some caveats.

Maybe I am being too skeptical about something I don't understand very well, but I'm not completely convinced it means no more kernel bugs will be found.

What does it mean for con leche if further lean kernel bugs are found? What if the bugs affect con leche's correctness itself?

chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
> Autumn of verified Lean kernels

Above my pay grade, but it sounds like some kind of super cool self-hosted recursive lean-checker-in-lean.

I notice there are a bunch of caveats in that part of the article. If you were an LLM strongly RL'ed to give humans the result they're asking for, using recursion bugs to satisfy the goal would probably be something you would try.

I'm by no means an expert but I do think the old adage "great claims require great evidence" still applies, and a healthy dose of scepticism and epistemic humility is warranted.

chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
> The existence of an undiscovered soundness bug doesn't make everything proven in Lean illicit. The proof would have to exploit the bug.

Yes, correct. The proofs would have to be checked to make sure they don't exploit the bug.

> computer formal verification could reach the trustworthiness of math itself.

Yes, it could. I don't think that is currently the case.

chr15m··on What mathematicians should know about the Lean Theorem Prover: reliability & AI
Will we ever see a soundness bug in the lean kernel again?

To a software developer the question seems insane. There were bugs in the past, of course there will be more.

When we see those bugs, what will it mean for AI lean proofs? Can we trust them?

This whole thing boils down to trust.

We can trust human verifications highly because of community and reputation and human proof-of-work. Humans sometimes lie about math results but it's rare because of this. They make mistakes and those mistakes are discovered by communities who are themselves largely trustworthy because of this.

LLMs don't care about reputation. They hallucinate and fabricate often. In harnesses they literally try to cheat and bend rules, which is a disaster for knowledge that is encoded in rules. So we have to rely on proof checkers.

The problem is, how trustworthy are proof checkers? Are there bugs? And is the underlying theory itself free of paradoxes and unknowables and mathematical "bugs" that can be exploited? Imagine a human mathematician hell bent on deceiving other mathematicians - would we trust their breakthroughs, even with a verifier?

Because of these properties, the final backstop has to be humans, and rooted in the community and proof-of-work based human trust system. At the moment people are trusting the tools too much.

Prediction: bugs will be found by humans using AI tools that call into question the Navier-Stokes proof.

chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
No worries!
chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
It is self hosted.

Ollama serves models on an OpenAI compatible API directly on the inference machine, which most open source harness software will connect to. The inference machine has a name like inferencemachine.local on my home network so the Ollama URL looks like http://inference machine:11434/v1 to connect to any model I've downloaded.

chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
We're like the proverbial kids in the candy store!
chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
Nextcloud for calendar and contacts. Android sync with DAVx.

The inference machine is a basic headless gaming PC with a second hand RTX3090 and 64GB RAM.

The solar panels clock 5kW which is mostly fed back into the grid. It's sunny a lot here. The maximum consumption of the PC is 700W and it's mostly idle. Broad strokes we're net positive.

chr15m··on OpenAI withdraws three mathematical results
> the mathlib has many gaps and lacks many basic theories. So many problems can not even be stated yet not to mention proved.

That is very useful context, thank you!

chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
Yep and the LLM can help explain, but it is actually super straightforward. Public key cryptography and the text file configs are tiny.
chr15m··on Ask HN: What do you run on a $5 VPS that's worth keeping online 24/7?
You could try with that as a prompt. I have a repo set up with all of my sysadmin scripts and so the LLM had that as additional context.
← PreviousPage 2 of 18Next →