Hacker News: inkysigma

New comment by inkysigma in "How LLMs work"

inkysigma — Sat, 06 Jun 2026 09:22:30 +0000

This is essentially an open research question. ML theory is unfortunately very weak relative to where the empirics are. I think there's a relatively optimistic paper that was posted a while back here but I would also take it with a grain of salt.

https://arxiv.org/abs/2604.21691

There's of course empirical results and relatively weak theoretical results like the UAT but I also don't think that answers your question fully, especially since it seems impossible to definitively answer questions that the industry seems to betting on like whether or not there is a lower bound to their error rate or whether hallucination as a problem can be solved. We have much stronger ideas of what linear regression is doing relative to what LLMs are doing.

New comment by inkysigma in "How LLMs work"

inkysigma — Sat, 06 Jun 2026 09:20:41 +0000

At a high level, the text samples are how the relationships are derived. If we treat text samples as sequences of tokens, then the sequences of tokens describe the joint distributions they occur together which confers the relationship between them. Iirc, this is related to the idea of the distributional hypothesis in NLP: the idea the semantics of words should be similar if they occur in similar situations.

New comment by inkysigma in "Everything in C is undefined behavior"

inkysigma — Wed, 20 May 2026 07:27:15 +0000

One example along this path as an example is that every function must either terminate or have a side effect. I don't think one has bitten me yet but I could completely see how you accidentally write some kind of infinite loop or recursion and the function gets deleted. Also, bonus points for tail recursion so this bug might only show up with a higher optimization level if during debug nothing hit the infinite loop.

New comment by inkysigma in "Bun is being ported from Zig to Rust"

inkysigma — Tue, 05 May 2026 01:26:43 +0000

So I can't tell if the linked commit is an actual attempt or just an experiment but it did always strike me as odd to make a JS runtime in Zig when my impression was there were a lot of work-stopping compiler bugs at the time.

New comment by inkysigma in "China blocks Meta's acquisition of AI startup Manus"

inkysigma — Mon, 27 Apr 2026 21:33:47 +0000

I'm curious how this view fits in with BERT or the T5 release which prior to the current LLM craze were the de facto language models for use in pretty much any tasks. Was this a position that would've otherwise grown without the llama release?

New comment by inkysigma in "A Boy That Cried Mythos: Verification Is Collapsing Trust in Anthropic"

inkysigma — Thu, 23 Apr 2026 08:41:52 +0000

Some of those organizations (Linux and Mozilla) work on open source code for which they are already trained on. For clients like Apple, they almost surely have agreements to not do that.

New comment by inkysigma in "Google broke its promise to me – now ICE has my data"

inkysigma — Wed, 15 Apr 2026 23:00:48 +0000

Well it did contain a request to not notify according to that same letter. I suppose that brings up several questions.

1. Does that mean the same thing in the ToS?

2. How valid are these requests?

New comment by inkysigma in "Google broke its promise to me – now ICE has my data"

inkysigma — Wed, 15 Apr 2026 22:40:54 +0000

Depends on how legitimate you consider an administrative warrant and how willingly you think complying with one is.

On a more practical level, forcing them to go to court might not be much better. If this went to a FISA court, those are essentially rubber stamps and give nearly 100% approval.

New comment by inkysigma in "EFF is leaving X"

inkysigma — Thu, 09 Apr 2026 18:20:22 +0000

His anti censorship stance isn't necessarily born out by the data:

https://www.washingtonpost.com/technology/2024/09/25/elon-mu...

New comment by inkysigma in "EFF is leaving X"

inkysigma — Thu, 09 Apr 2026 18:18:59 +0000

X under Musk has sustained more government takedown requests.

https://www.washingtonpost.com/technology/2024/09/25/elon-mu...

New comment by inkysigma in "Epoch confirms GPT5.4 Pro solved a frontier math open problem"

inkysigma — Tue, 24 Mar 2026 03:21:56 +0000

I think in this context, scaffolds are generally the harness that surrounds the actual model. For example, any tools, ways to lay out tasks, or auto-critiquing methods.

I think there's quite a bit of variance in model performance depending on the scaffold so comparisons are always a bit murky.

New comment by inkysigma in "Illinois Introducing Operating System Account Age Bill"

inkysigma — Tue, 17 Mar 2026 19:04:34 +0000

Actually given boot chain protection, this will probably get harder as time goes on but even assuming some kids are able to, this is clearly definable as a user error: the fault lies with the kid and as a parent you need to think about your threat model.

Right now, it's not even clear how to create parental controls at a reasonable level so there's no clear path for what to do or how to respond.

New comment by inkysigma in "Illinois Introducing Operating System Account Age Bill"

inkysigma — Tue, 17 Mar 2026 18:54:09 +0000

Except none of these bills (California or the one in question) as currently written require an ID to actually be verified, merely that the user provide an age. This seems intentional as it's seems to solve the user journey where a parent is able to set a reasonable default by simply setting up an associated account age at account creation. It's effectively just standardizing parental controls.

I think this is a reasonable balance without being invasive as there's now a defined path to do reasonable parenting without being a sysadmin and operators cannot claim ignorance because the user input a random birthday. The information leaked is also fairly minimal so even assuming ads are using that as signal, it doesn't add too many bits to tracking compared to everything else. I think the California bill needs a bit of work to clarify what exactly this applies to (e.g. exclude servers) but I also think this is a reasonable framework to satisfy this debate.

I've seen the argument that this could lead to actual age verification but I think that's a line that's clearly definable and could be fought separately.

New comment by inkysigma in "Age Verification Lobbying: Dark Money, Model Legislation, Institutional Capture"

inkysigma — Sat, 14 Mar 2026 09:45:08 +0000

I think this is the third time this has effectively been posted see:

https://news.ycombinator.com/item?id=47362528

https://news.ycombinator.com/item?id=47365597

New comment by inkysigma in "Meta Platforms: Lobbying, dark money, and the App Store Accountability Act"

inkysigma — Fri, 13 Mar 2026 19:14:57 +0000

Some of these are also just like really weak? One of them for example seems to be some random employee at FB donating ~$1k to a politician and calling that a link. The entire "Proven Findings" is all over the place and provides no coherence. I don't think it's a particular secret that Meta would prefer age verification be done at the OS level so I'm not really sure what the added claim here is.

> A Meta employee (Jake Levine, Product Manager) contributed $1,175 to ASAA sponsor Matt Ball's campaign apparatus on June 2, 2025. Source: Colorado TRACER bulk data.

> No direct Meta PAC contributions to any ASAA sponsor across Utah, Louisiana, Texas, or Colorado. Source: FollowTheMoney.org multi-state search.

While it is true that Meta has funded groups that advocate for age verification, a lot of them also appear to have other actors so it's not like this is some pure Meta thing as some of the other commenters are suggesting.

New comment by inkysigma in "I'm reluctant to verify my identity or age for any online services"

inkysigma — Tue, 03 Mar 2026 20:07:03 +0000

Depending on the implementation, I could see that having rate limiting effects. There're only finitely many IDs so scaling sockpuppeting will saturate these IDs quickly but it's quite easy to spin up a new anonymous account. For example, I think the EU ID system has an upcoming way to create pseudo anonymous identifiers that can identify a user per website.

This presents the problem of governments being able to gatekeep speech which I am quite uncomfortable with but maybe there's some safeguard within the eIDAS proposal that makes this idea incorrect?

New comment by inkysigma in "I'm losing the SEO battle for my own open source project"

inkysigma — Tue, 03 Mar 2026 19:22:49 +0000

Just an FYI, but I don't know if being in the website field of GitHub really helps since there's a rel nofollow on the link.

New comment by inkysigma in "We Will Be Divided"

inkysigma — Sat, 28 Feb 2026 10:49:10 +0000

I mean in the sense that they seem to literally be random names. I don't even think they're people associated with Palantir in anyway.

New comment by inkysigma in "We Will Be Divided"

inkysigma — Sat, 28 Feb 2026 10:44:21 +0000

I'm assuming this is satire but I'm wondering why include names of seemingly random people? Why not leave it empty or make it signed by high level known executives.

New comment by inkysigma in "The United States and Israel have launched a major attack on Iran"

inkysigma — Sat, 28 Feb 2026 08:45:39 +0000

Once that side channel was found, it was kind of inevitable it would be plugged. Even under a normal administration, that's an opsec leak.