Hacker News Reader

fwlron Typesafe AI raises $870M at $7.5B

Others have capably pointed out why their technology cannot be much of this valuation. I submit that “the ability to grab attention in the AI room” is what is being valued this highly.

fwlron Computers Cannot Make Decisions

What kind of test can you imagine where confirmatory evidence would not also be evidence of misconduct opening the company up to legal liability?

fwlron Eye of Sauron: Long-Range Hidden Spy Camera Detection (2024)

Oh, that is fascinating - I think a scheme like that could very well give you constant-space as well as constant-time, while calculating and preserving the desired compression information for use in transmission (which itself could separately be made constant for detectable channels).

fwlron Computers Cannot Make Decisions

If you brought down prod and deleted customer data and told them Claude did it, you’d still be the one in trouble. So they do know how to hold people accountable for actions taken by AI.

fwlron Eye of Sauron: Long-Range Hidden Spy Camera Detection (2024)

Potentially, although raw pixel data presents its own challenges - one 1080p60 camera will fill a terabyte of storage in less than an hour.

A better mitigation might be to take a leaf from cryptography and develop a constant-time codec.

fwlron Scam American companies are using to manipulate ingredient lists

Sorry if it sucks for their competitive advantage or profit margin or whatever but people are literally going to put this stuff in their bodies to sustain their existence so you have to say what’s in it. If you want to keep your product composition a mystery go make SaaS and put it behind an API.

fwlron REA Reverse – Engineer Anything

Probably it is not better. I think this is aimed at people who do not have a “current RE methodology”, do not know enough to specify things like Ghidra, etc., but who do have a desire to feel like they reverse-engineered and can reliably predict that a conversation with a chatbot will make them feel that way.

fwlron OpenAI, the Partition Principle, and Mathematics

Dredging every inch of the lake to catch every single fish and dumping them all in one big pile the town square is just “solving fishing and releasing the fish to the public”, this anti-trawler hysteria is getting really funny.

fwlron OpenAI, the Partition Principle, and Mathematics

I don’t get why you aren’t upset. In general I am actually quite disappointed by how meekly the software engineers handed our industry over to AI. But I am particularly revolted by those who exercise their free will to rise from the foetid swamp and say “come on in, the water’s great!”.

fwlron “Math 2.0” will need to value mathematical progress more holistically

Potentially years, for the most important problems with the densest thickets of agentic reasoning to cut through! But if that is what it takes, then that is what it takes, it’s not possible to win faster. Dumping it early is not winning faster, it is losing, specifically by forfeiting and tossing the table so no one else can win either. Very unsportsmanlike.

And I do not buy that it is “unfair” either. That argument proves far too much - it would equally apply to peer review, editing, revising, publication, and indeed any non-essential activity by the author from the moment of their epiphany onward.

Note also that my suggestion is something I came up with on the spot in a handful of seconds for a field I am entirely unfamiliar with. It should be trivial in the extreme for a well-compensated and motivated professional familiar with the field to take a few weeks and vastly improve on it.

fwlron “Math 2.0” will need to value mathematical progress more holistically

Dumping the one proof they have access to, sticking their name next to “OpenAI” and having it preside over reams of incomprehensible slop? Instead of taking a few months to digest, improve, rework it? I mean, maybe a few would, but I don’t think this objection holds much water.

fwlron AHM Statement on OpenAI's October 6 Release of Mathematical Documents

As I understand it, this is one of the groups OpenAI claimed to be consulting with to improve their process after the Navier Stokes fumble. I think OpenAI might have even said something like “in accordance with this group’s guidance” in the big dump? So this post should be read as distancing themselves from OpenAI and denying any legitimacy OpenAI tried to derive from their association.

Edit: on researching more, I don’t think this is the case; I must have confused them with some other group.

fwlron “Math 2.0” will need to value mathematical progress more holistically

For each paper they think they have, reach out to mathematicians working in the relevant area to find ones willing to take on the role of advisor, co-author, reviewers, etc. Basically just follow the same process that a postdoc would to publish a paper.

fwlron Will AI kill us all

    Every number below is someone's judgment, and the answer changes with who is asked

Ah, yes, different people have different answers. Insightful as always, Claude.

fwlron “Math 2.0” will need to value mathematical progress more holistically

Wouldn’t be the first time someone kept the bean counters happy while also generating some positive externalities! Of course, now with AI we might finally be able to realise maximum efficiency, where only the bean counters are happy and there’s no other positives.

fwlron “Math 2.0” will need to value mathematical progress more holistically

Well, that’s the thing, isn’t it? They somehow invented a way of solving an open problem without advancing the field. That’s something that mathematicians had never even thought was possible. And while the mathematicians were beginning the great conversation to re-examine their fundamental understanding of the pursuit of math in light of this new phenomenon, OpenAI decided they would do it again 372 more times.

fwlron “Math 2.0” will need to value mathematical progress more holistically

I hope Terry has somewhere private where he can safely express his anger at how poorly math has been treated by these AI corps. I understand that as a recently pro-AI public figure he is limited to ambivalence, so I understand him adding caveats like “maybe math 2.0 has a place for AI”, but it can’t feel good to say stuff like that just days after OpenAI so drastically salted the earth.

fwlron Port of the TypeScript compiler, checker and lsp to Rust, by LLM

    # “The Slop Line"

      Everything below this was written by my LLMs, not me.

I really appreciate this convention, thank you.

fwlron Erdosproblems.com Succumbs to the AI Onslaught

Within the heterogeneous group “programmers”, those who are benefiting the most from AI are also among the least scrupled (see e.g. Digital Ocean can find $3,000,000.00 in 4 days for DHH’s vibe-coded distro setup scripts[1]), while the most selfless are getting shafted the hardest (see e.g. the same Digital Ocean can’t find $50.00 a month for the open-source developers who build and maintain the distros[1]).

1. https://keith.is/post/digitalocean-and-css-tricks/

fwlron When random is not actually random enough

I don’t think the “random uint” api is too low-level, or lacks a pit of success - I think you’re just reaching for the wrong api. The problem of “make n bits pseudo randomly set to either 1 or 0” is nearby to your problem of “choose an element according to a probability distribution”, but it’s a separate problem in its own right.

I think actually this is an argument for language designers to include a “std.choice” in their standard library that consumes random bytes and correctly performs common ergonomic operations like “get one element at random from this collection”.

(If your standard library tries to make a distinction between “regular random number generators” and “cryptographically secured random number generators”, I think this distinction between “generate random bits” and “make probabilistic choices” is about equally important.)

fwlron A sustainable web career, for when all this blows over

What sphinx of cement and aluminum…

fwlron Erdosproblems.com Succumbs to the AI Onslaught

    Of all sad words of tongue or pen,
    the saddest are these:
    a tragedy of the commons has happened, again.

fwlron Erdosproblems.com Succumbs to the AI Onslaught

It is becoming exceedingly difficult to ignore that the benefits of AI accrue disproportionately to those most deficient of scruples, while its costs and harms fall upon the selfless and pleasant like laser-guided virtue-seeking munitions.

I do not particularly care to debate a theory of why it is so, nor press a charge of malice or intent; mere observation suffices to convince me that this shit is corrosive to the better nature of humanity.

fwlron Samon: Designing a Zen Garden Raking Puzzle

For what it’s worth, I too was a little surprised and disappointed by the quality of the prose, until I double-checked the url and realised this was more like “a design doc artifact hosted on gwern.net” than “an article posted on gwern.net”.

fwlron A single license fee of £100k.00

Move fast, break things. If AI had waited until the law caught up, we’d still have Clippy correcting our typos instead of Claude yolo-ing our entire job.

fwlron Demystifying Tufte's data-ink ratio

Mistifying Tufte’s data-ink ratio (by blasting it into a fine mist with my slop cannon).

fwlron Accept 'bad things' in return for benefits of AI, says Sam Altman

OpenFlame’s latest model has really good scores on FahrenBench but I’m having a lot of trouble keeping it contained. I use a proper firepit but it keeps spreading to nearby homes and vehicles.

fwlron Anthropic reported diary entry to police, woman faces felony charge

Both of those scenarios would demonstrate even more commitment to safety so I imagine they’d be at least as likely to happen as this, if not more.

fwlron Google freezes open-source bug bounty program amid flood of invalid AI slop

Funniest shit I’ve read all day

fwlron The evolution of effective altruism

What’s that thing pg said about a submarine? Somebody’s got a dossier of all the weirdest things about effective altruism they can find (not that it’s in short supply), and they’re out there convincing reporters who need to run articles about AI that this is the ideology of the AI guys, and wouldn’t your readers like to know all the lurid details?

While it’s probably something prosaic as “Anthropic pays the most lip service to EA, this makes them look bad” or “Our employees are leaning EA and it’s causing problems, things would be easier if they were too embarrassed to admit they were EA”, I do wonder - the board member who instigated the firing of Sam Altman from OpenAI does happen to be a long-time effective altruist…