Hacker News: nsagent

Research Shows Verbatim Recall of Copyrighted Books in LLMs

nsagent — Thu, 26 Mar 2026 04:28:22 +0000

Article URL: https://cauchy221.github.io/Alignment-Whack-a-Mole/

Comments URL: https://news.ycombinator.com/item?id=47526640

Points: 4

# Comments: 0

New comment by nsagent in "Android introduces $2-4 install fee and 10–20% cut for US external content links"

nsagent — Sat, 20 Dec 2025 05:29:37 +0000

Hopefully this gets slapped down hard just like Apple recently did. Both Apple and Google want to continue business as usual despite the court rulings.

New comment by nsagent in "Amazon will allow ePub and PDF downloads for DRM-free eBooks"

nsagent — Fri, 19 Dec 2025 17:20:36 +0000

They screwed me in a different way. I simply didn't log into Amazon for a couple years as I've tried to minimize my use of Amazon. When I went to log in, they locked my account without any way to unlock it. Talking with support multiple times did nothing. Now all my digital purchases are gone.

Edit: If anyone knows a way to get them to unlock the account, I'd appreciate it. They won't issue a password reset or anything similar, which seems ridiculous considering they never claimed fraud. Simply that it had been too long since I logged in.

New comment by nsagent in "Ask HN: Does anyone understand how Hacker News works?"

nsagent — Fri, 19 Dec 2025 02:21:14 +0000

[flagged]

New comment by nsagent in "Ask HN: Does anyone understand how Hacker News works?"

nsagent — Fri, 19 Dec 2025 01:29:57 +0000

I'll be honest and blunt: I try to avoid his blog posts and comments as much as possible because I do find his contributions to be super spammy. It's about the frequency of his self-promotion (as a researcher in NLP it's tiresome to constantly see his self-promotion on nearly all posts on HN related to LLMs). Seems I'm not the only one.

If he posted on the ML subreddit while I was still a mod there (left after the API kerfuffle) I would have messaged him and asked him politely to tone it down.

New comment by nsagent in "Coursera to combine with Udemy"

nsagent — Thu, 18 Dec 2025 19:49:03 +0000

Agreed. That's how I got my start in ML, through Andrew Ng's course a little over a decade ago. I then went on to get my PhD in CS focused on NLP.

New comment by nsagent in "All OpenReview Data Leaks"

nsagent — Thu, 27 Nov 2025 21:43:38 +0000

https://xcancel.com/iclr_conf/status/1994104147373903893

https://safereddit.com/r/PhD/comments/1p85o3j/openreview_bug...

https://safereddit.com/r/MachineLearning/comments/1p85vs0/d_...

All OpenReview Data Leaks

nsagent — Thu, 27 Nov 2025 21:42:05 +0000

Article URL: https://twitter.com/iclr_conf/status/1994104147373903893

Comments URL: https://news.ycombinator.com/item?id=46073488

Points: 6

# Comments: 1

New comment by nsagent in "AI has a deep understanding of how this code works"

nsagent — Thu, 27 Nov 2025 07:57:22 +0000

Oh wow. That's just egregious. Considering the widespread use of Auth0, I'm surprised this isn't a bigger story.

New comment by nsagent in "AI has a deep understanding of how this code works"

nsagent — Tue, 25 Nov 2025 19:58:18 +0000

It's clear Claude adapted code directly from the OxCaml implementation (the PR author said he pointed Claude at that code [1] and then provides a ChatGPT analysis [2] that really highlights the plagiarism, but ultimately comes to the conclusion that it isn't plagiarized).

Either that highlights someone who is incompetent or they are willfully being blasé. Neither bodes well for contributing code while respecting copyright (though mixing and matching code on your own private repo that isn't distributed in source or binary form seems reasonable to me).

[1]: https://github.com/ocaml/ocaml/pull/14369#issuecomment-35573...

[2]: https://github.com/ocaml/ocaml/pull/14369#issuecomment-35566...

New comment by nsagent in "Git 3.0 will use main as the default branch"

nsagent — Mon, 24 Nov 2025 07:51:09 +0000

Because it goes hand-in-hand with the euphemism treadmill. I defer to the late great George Carlin's words on the subject:

  I don't like euphemisms, or euphemistic language. And American English is loaded with euphemisms. Cause Americans have a lot of trouble dealing with reality. Americans have trouble facing the truth, so they invent the kind of a soft language to protest themselves from it, and it gets worse with every generation.

Video https://www.youtube.com/watch?v=vuEQixrBKCc or text form: https://www.lingq.com/en/learn-english-online/courses/87644/...

New comment by nsagent in "Homeschooling hits record numbers"

nsagent — Sat, 22 Nov 2025 07:23:05 +0000

As a bright student who was never challenged in K-12, I can unequivocally state that this ultimately hurt me in the long run. I seriously didn't know how to study and didn't care to try learning when I actually needed it in some of my undergrad courses.

For example, when I took trigonometry in high school I did none of the homework, showed up to the tests and aced them. That led me to getting a C in that class (kindly the teacher advanced me to pre-calc, but forced me to retake trig as well). That's basically the attitude I had throughout high school and undergrad. I'm positive I could have amounted to more earlier in life (only years later did I return to academia to earn my PhD in CS after tiring of industry).

New comment by nsagent in "Open-source Zig book"

nsagent — Mon, 17 Nov 2025 12:39:43 +0000

I literally just came across this resource a couple of days ago and was going to go through it this week as a way to get up to speed on Zig. Glad this popped up on HN so I can avoid the AI hallucinations steering me off track.

New comment by nsagent in "Report: Tim Cook could step down as Apple CEO 'as soon as next year'"

nsagent — Sun, 16 Nov 2025 22:12:15 +0000

> Ideas are easy.

Let's not get into absolutes. Ideas can be as hard or harder than execution. Not many will claim ideas like Special Relativity are easy to come by.

It really depends on the idea and what you're executing. Often in business both the idea and the execution have to work out to captialize. In this case Jobs might have set the direction and Cook capitalized. Both deserve a share of the credit.

New comment by nsagent in "wBlock: The next-generation ad blocker for Safari"

nsagent — Tue, 11 Nov 2025 22:04:32 +0000

I was previously using Wipr 2, but got tired of the fact that it would block things I didn't want and there was no recourse except disabling it wholesale (for example, I noticed it blocks any element with an id of 'privacy_policy' or 'privacypolicy,' which I ran into when actually trying to read the privacy policy for a service I use).

wBlock fixes that by allowing you to have custom rules. It also includes the ability to run userscripts, so I no longer need a separate extension for that.

wBlock: The next-generation ad blocker for Safari

nsagent — Tue, 11 Nov 2025 21:58:26 +0000

Article URL: https://apps.apple.com/us/app/wblock/id6746388723

Comments URL: https://news.ycombinator.com/item?id=45893452

Points: 2

# Comments: 1

Recommended Practices for NPOV Research on Wikipedia

nsagent — Wed, 05 Nov 2025 22:33:26 +0000

Article URL: https://arxiv.org/abs/2510.21526

Comments URL: https://news.ycombinator.com/item?id=45828998

Points: 2

# Comments: 0

New comment by nsagent in "The Case That A.I. Is Thinking"

nsagent — Tue, 04 Nov 2025 19:16:22 +0000

You might be interested in reading about the minimum description length (MDL) principle [1]. Despite all the dissenters to your argument, what your positing is quite similar to MDL. It's how you can fairly compare models (I did some research in this area for LLMs during my PhD).

Simply put, to compare models, you describe both the model and training data using a code (usual reported as number of bits). The trained model that represents the data within the fewest number of bits is the more powerful model.

This paper [2] from ICML 2021 shows a practical approach for attempting to estimate MDL for NLP models applied to text datasets.

[1]: http://www.modelselection.org/mdl/

[2]: https://proceedings.mlr.press/v139/perez21a.html

New comment by nsagent in "State of Terminal Emulators in 2025: The Errant Champions"

nsagent — Tue, 04 Nov 2025 14:13:50 +0000

Maybe you mean it's too much effort, because I'm sure you could. I was taught touch typing on a QWERTY keyboard in the summer between 6th and 7th grade. Last year I switched to Colemak after nearly 30 years of QWERTY.

New comment by nsagent in "Updated practice for review articles and position papers in ArXiv CS category"

nsagent — Sat, 01 Nov 2025 22:54:59 +0000

  To every man is given the key to the gates of heaven. The same key opens the gates of hell.

-Richard Feynman

https://www.goodreads.com/quotes/421467-to-every-man-is-give...

https://calteches.library.caltech.edu/1575/1/Science.pdf