Hacker Newsnew | past | comments | ask | show | jobs | submit | ziofill's commentslogin

I don’t think this is true.

Could you please explain why? It's an additive update to the weights, adding an outer product of a vector with its transpose, which must have rank 1. What am I missing?

This comment is hysteresical

I find it polarizing.-

It can be quite eddyfying

A harmonious comment about an impedance mismatch quickly spiraled into a galvanizing flow

It's just that HN is currently saturated with this kind of thing

Regardless of flux, we still haven't addressed the core issue.

Come on now, I don’t think it was meant as a justification of that inexcusable driver’s behaviour.


What was the comment trying to say?


the downvoted comment sounded like textbook whataboutism to me. those who responded have their antennae up for rhetorical fallacies.

https://en.wikipedia.org/wiki/Whataboutism


Classifying something as a textbook whataboutism is typically itself a rhetorical fallacy. It is rare for it to be used in the sense of identifying a real problem. Normally when it gets invoked someone has an argument, often a reasonable one, and someone else wants to pretend it isn't a valid but can't come up with a logical inconsistency.

In this case nxm's comment has formal problems - it is just irrelevant. True or not, it doesn't justify mowing down cyclists.


As cool as this looks, I can’t get past my suspicion of Meta…


I’m a professional theoretical physicist in my 40s and I never thought the day would come that I would realize I became shit at mental math. I can do quantum physics in my head but don’t ask me to add 17 to 34


Thanks for mentioning this. It makes me feel better about stumbling with metal math in front of some colleagues that are faster.


I've also noticed that I drive more slowly/defensively, if only because I need more time to react to all these other bad drivers =P

You won't catch me (nor any of my brilliant brothers) doing math without a calculator... and I got a SAT790Math way-back-when (do they still score that way? I missed one question, only; many of my brothers scored 750+).


I’ve tried to keep up with “best practices” around AI use, but things are improving so quickly that I’ve largely given up. The vanilla agents are just fine for my needs as they come.


Can you elaborate more how do you setup vanilla agents ? Which agents you use and which use case that it’s greatly show benefit for you. Thanks


That’s what I’m saying, I don’t set it up. I just install Claude code and I’m good to go.

Managing configuration files feels like worrying about fine-tuning in 2023


I don’t think so: the ELO rating difference is what matters, e.g. a 1000 player vs an 800 player has the same win probability as a 2800 vs a 2600.


George Costanza would be proud


That’s a fair question, but it seems that it’s not yet necessary. See here

https://dylancastillo.co/posts/pelicanmaxxing.html

https://simonwillison.net/2026/Jul/22/


I don't think this argument is a good one though, as it would be quite natural for a lab rhat want to macimize the performance of their model on the pelican bench to train it for “text-to-svg simple image generation” rather than just “pelicans on bicycle”.


That is the point though, if labs are maximizing svg image generation capabilities it is a very good thing. That's a general skill that is useful. So assuming they aren't specifically maximizing pelican bicycle svgs (and it doesn't look like they are) then incentives are aligned that the "benchmark" is measuring a general desirable capability.


Gemini have done exactly that.

(I doubt it's because of my stupid benchmark, though!)



Reminded me of this https://xkcd.com/1367


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: