HNHacker News
TopNewBestAskShowJobs

chunky1994

1,431 karma · joined May 23, 2011

2x Startup founder (exited). University (of Waterloo) grad in mathematical physics, interested in anything that makes me curious.

I'm on the lookout for some hard, and interesting things to work on.

submissionscomments
chunky1994··on More questions about whether researchers can trust OpenAI with unpublished math
Why are we being so charitable to trillion dollar organizations here? If OAI keeps re-enabling the train model toggle on every app update to codex, does it also fall under "common sense opsec" to re-disable this toggle every time?

Arguably you expect that unless you are explicit about providing permissions to these labs to use your data for training then your data is yours, and not theirs. Especially on a paid account (let alone an enterprise one). Why is the opt-out supposed to be "common sense opsec" rather than the opt-in should be common sense regulation?

chunky1994··on List of references on Sony websites to players "owning" their digital games
The case wasn't dismissed because of the arbitration clause. Disney dropped the argument after they got severely lambasted in public over the absurdity of the defense. It was dismissed later by the plaintiff (so there was likely some settlement)
chunky1994··on The Doorman's Fallacy in action
Yes, this is quite standard outside the US. In Canada, Mexico, Europe, Asia etc. this is more the standard practice than the opposite.
chunky1994··on Show HN: Learn from 30 historical figures, open source, nonprofit, self-hosted
Why though?

We have writing, artifacts and objects from ancient peoples which we then use to try to construct historiographies of those cultures, as well as interpretations of their lived experience and circumstances.

This is just doing it for specific historical figures with a different type of technology. Why is it more disrespectful than what historians do?

chunky1994··on AI-assisted cognition endangers human development?
Does anyone use LLMs in such a manner that they believe it always has the most up to date information (without web search tools?).

Isn't this whole thesis negated by the fact that tool calling web search exists? This just feels like a whole lot of words to say, don't treat a LLM as an always up to date infallible statistical predictor.

chunky1994··on Mystery jump in oil trading ahead of Trump post draws scrutiny
@dang, this has way more discussion than the previous threa,d but people can't see this because it's a dupe.
chunky1994··on Tell HN: Litellm 1.82.7 and 1.82.8 on PyPI are compromised
Their Personal Access Token must’ve been pwned too, not sure through what mechanism though
chunky1994··on Major insider trading on oil detected ahead of Iran talks
Realistically I think it will come down to the aggrieved counterparties here. Who was on the losing side of the money, was it Joe Schmoe day trader or a bunch of funds who lost their shirt?

If it’s the hedge funds or institutional money, you can absolutely be sure this will come to a head. People don’t like being taken for a ride, and if they are repeatedly taken for a ride and they are organized market participants they will come around and make sure there is a comeuppance as a collective

chunky1994··on The United States and Israel have launched a major attack on Iran
Fair point, but the IRGC telling ships to turn around, as opposed to the ships themselves doing it (as per reporting) would imply that the Strait has been blockaded in some fashion. It remains to be seen if this is all a bluff, I'm just as skeptical as this would be their last option, but given the strikes on other Gulf countries, the threat seems a bit more plausible of actually being real.
chunky1994··on The United States and Israel have launched a major attack on Iran
There is credible reporting (Reuters etc.) that ships are being turned around, so it does appear that the mines (or at least threat thereof) have been deployed. Either way, as long as the threat of sinking is alive the strait is uninsurable and is for all practical purposes closed.
chunky1994··on Management as AI superpower: Thriving in a world of agentic AI
> The dark side of this same coin is when teams try to rely on the AI to write the real code, too, and then blame the AI when something goes wrong. You have to draw a very clear line between AI-driven prototyping and developer-driven code that developers must own. I think this article misses the mark on that by framing everything as a decision to DIY or delegate to AI. The real AI-assisted successes I see have developers driving with AI as an assistant on the side, not the other way around. I could see how an MBA class could come to believe that AI is going to do the jobs instead of developers, though, as it's easy to look at these rapid LLM prototypes and think that production ready code is just a few prompts away.

This is what's missing in most teams. There's a bright line between throwaway almost fully vibe-coded, cursorily architected features on a product and designing a scalable production-ready product and building it. I don't need a mental model of how to build a prototype, I absolutely need one for something I'm putting in production that is expected to scale, and where failures are acceptable but failure modes need to be known.

Almost everyone misses this in going the whole AI hog, or in going the no-AI hog.

Once I build a good mental model of how my service should work and design it properly, all the scaffolding is much easier to outsource, and that's a speed up but I still own the code because I know what everything does and my changes to the product are well thought out. For throw-away prototypes its 5x this output because the hard part of actually thinking the problem through doesn't really matter its just about getting everyone to agree on one direction of output.

chunky1994··on This is not the future
But the world is not deterministic, inherently so. We know it's probabilistic at least at small enough scales. Most hidden variable theories have been disproven, and to the best of our current understanding the laws of the physical universe are probabsilitic in nature (i.e the Standard Model). So while we can probably come up with a very good probabilistic model of things that can happen, there is no perfect prediction, or rather, there cannot be
chunky1994··on Physicists Take the Imaginary Numbers Out of Quantum Mechanics
Dummit and Foote is the classic abstract Algebra textbook to learn about how to precisely define these. Its treatment of ring theory is very well motivated and easy to grasp
chunky1994··on Port of Los Angeles says shipping volume will plummet 35% next week
I don't think anyone is advocating for incentivizing forced/child labour.

Given that the ILAB link you posted itself is maintained by EO 13126 signed by the Clinton Administration, I think there can be nuance in the discussion around whether or not the blanket application of certain foreign policy instruments is the right way to induce a change in the domestic policy of another country to solve the problem of bad labour practices.

We can do this without it becoming an argument about whether trade is "good" or "bad" depending on what "side" you are on.

chunky1994··on Port of Los Angeles says shipping volume will plummet 35% next week
This is a good discussion around the supply chain issues that will likely be happening: https://youtu.be/-dgHWv-Dh6Q?t=1370

Ryan runs Flexport which is a supply chain company so its from the "source" if you will.

chunky1994··on I fear for the unauthenticated web
This difference in emotional reaction is because of the effort involved in the process. Functionally, we see YouTube video creation as a fundamentally difficult exercise (to do well) and results in a singular product (one video). Any additional content would need an ongoing investment of time and money from the creator. The LLMs though would not require an ongoing investment beyond the first training run, that is probably why you have a problem with it, they're an extremely high leverage way of taking advantage of content.
chunky1994··on Promising results from DeepSeek R1 for code
The difference is though there isn't a whole lot of "whole cloth novel solutions" being written in software today so much as a "write me this CRUD app to do ABC" which current generations are exceedingly good at.

There are probably 10% of truly novel problems out there, the rest are just already solved problems with slightly different constraints of resources ($), quality (read: reliability) and time. If LLMs get good enough at generating a field of solutions that minimize those three for any given problem, it will naturally tend to change the nature of most software being written today.

chunky1994··on Weight-loss drug found to shrink heart muscle in mice, human cells
I'd like to see the diets in the study that are specified as the "calorie-reduced diets". (Can't seem to find the paper). If it's the same as the Standard American Diet, this muscle loss is quite explainable. I think the mitigation is relatively easy though, if you want to shift the p-ratio, recommending a daily high protein shake would do a lot to stave off muscle loss (and even more if resistance training is applied of course). The exercise addition is probably the hardest to adhere to.
chunky1994··on How to Reach Angel Investors in the San Francisco Area for Agentic AI Startup?
I'd suggest investing your time in the product first and converting the waitlist to real users before trying to take on external investment. It will either make it much easier or you won't need to raise depending on what happens when your product goes out to the 1000 user list.
chunky1994··on What Kind of Writer Is ChatGPT?
If you train one of the larger models on these specific problems (i.e DM for D&D problems) it probably will surprise you. The larger models are great at generic text production but when fine-tuned for specific people/task emulation they're quite surprisingly good.
chunky1994··on Where is Noether's principle in machine learning?
I'm a bit skeptical to give up conservation of energy in a system with friction. Isn't it more accurate to say that if we were to calculate every specific interaction we'd still end up having conservation of energy. Now whether or not we're dealing with a closed system etc becomes important but if we were to able to truly model the entire physical system with friction, we'd still adhere to our conservation laws.

So they are not approximations, but are just terribly difficult calculations, no?

Maybe I'm misunderstanding your point, but this should be true regardless of our philosophy of physics correct?

chunky1994··on Where is Noether's principle in machine learning?
The conserved quantity is derived from Noether's theorem itself. One thing that is a bit hairy is that Noether's theorem only applies to a continuous, smooth (physical -> there is some wiggle room here) space.

When deriving the conservation of energy from Noether's theorem you basically say that your Lagrangian (which is just a set of equations that describes a physical system) is invariant over time. When you do that you automatically get that energy is conserved. Each invariant produces a conserved quantity as explained in parent comment when you apple a specific transformation that is supposed to not change the system (i.e remain invariant).

Now in doing this you're also invoking the principle of least action (by using Lagrangians to describe the state of a physical system) but that is a separate topic.

chunky1994··on Where is Noether's principle in machine learning?
Right, the rephrasing of the sentence is a tad more accurate. Your three entities are [invariant -> conserved quantity]: (translation -> momentum), (rotation -> angular momentum) and (time -> energy).
chunky1994··on How ChatGPT made me lazy
This is exactly why using GPT to supplement but not supplant writing code is very good. Code generation is generally more helpful as you go to higher layers of abstraction as long as you can stay at that layer of abstraction. Once you need to jump a level lower because your abstraction fails in a specific context you need pre-existing knowledge of how the abstraction works in order to ask GPT the right questions.

Hence, depending on the domain knowledge of the piece of code you are dealing with, GPT can be very helpful in generating a good scaffolding to get off the ground quickly (i.e you want to write a web app but don't want to deal with having to learn how to write a whole react app etc.), but asking GPT to...write an optimizer for a C runtime would end up with poor results as its heavily bent on the specifics of that task where a specialists' knowledge would outweigh any abstractive advantages.

One very useful experiment I did early on was try to solve a problem with GPT where I had deep domain expertise and see where the cracks are vs in one in which I had very poor expertise. This led me to make my abstraction based statement above, and so far I've seen it remain true with every successive version.

chunky1994··on Framing Frames: Bypassing wi-fi encryption by manipulating transmit queues
So the way it works is that if you have a setup like so: Your client <-- bad actor --> Public Webserver.

Your first layer of defense is assuming its hard for a bad actor to get to that spot (which this attack is showing is possible).

Now the next thing the bad actor needs to do to compromise your traffic is deal with HTTPS/VPN. The only real way to do that would have to be to convince your client that the certificate from the CA that's signing the https traffic received from the webserver/VPN provider is trusted/ (i.e a classic MITM with a compromised client and a local CA).

Most clients will warn you that this cert is unrecognized, so with an untampered client unless you just clickthrough the cert warnings you will be protected. However, people often ignore cert warnings and that means that all your traffic is now cleartext for the bad actor.

chunky1994··on Framing Frames: Bypassing wi-fi encryption by manipulating transmit queues
I'd argue the number of startups that have trust all local network connections is not far behind. I doubt a company seriously considers their wifi connectivity as a threat vector until they get to a series B size.

Hypothetically I'd imagine if you take early startup infrastructure snooping + <insert criminal act here> you probably have a criminal enterprise that's cashflow positive.

chunky1994··on macOS Sonoma Forced Installs
I am assuming this is a bug, otherwise it would be a very weird move on apple's part. While they're usually pushy about updates etc. for anyone who uses defaults they usually respect user decisions.

I've dealt with the apple auth chain quite a bit and from the reports it looks like the earliest this started happening was Jan 10th with this notification "macOSInstallerNotification_RC" getting pushed to everyone on Ventura. Very likely the bug then make this appear as a required security update and auto installs it despite user input (I am not sure how many people without the install security updates toggle have also been upgraded).

Also if you have a firmware password on that is still needed for security upgrades so that's probably why there's a difference between password less updates and non-passwordless upgrades but that's just a guess.

chunky1994··on Audiobox: Meta's new foundation research model for audio generation
I used to be an LLM until I took an arrow to the knee. Aside from the joke, I think the barrier would definitely reduce and in-game characters would be contextually far more aware but how do you enforce plot progression in such a truly open world? Can you control the boundary of LLM expression?
chunky1994··on Sam Altman is still trying to return as OpenAI CEO
Correlation =/= causation. This is most likely coincidental. I highly doubt Dustin's differing views caused a (near) unanimous ousting of a completely different company's CEO that had nothing to do with Dustin's primary business.
chunky1994··on Ask HN: How will the $900K median OpenAI salary work if employed at Microsoft?
It's 300k cash, rest in "profit participation". MSFT has enough RSUs to make that happen without it being a problem
← PreviousPage 2 of 9Next →