Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Prism

openai.com

Post reply on HN
Print view View original

411–420 of 543 posts

Re: Prism

#411

Post by lionkor »

doodlesdev wrote: post #312
ivirshup wrote: post #264

Earlier quoted context omitted.

I've heard that now that AI conferences are starting to check for hallucinated references, rejection rates are going up significantly. See also the Neurips hallucinated references kerfuffle [1] [1]: https://statmodeling.stat.columbia.edu/2026/01/26/machine-le...

Honestly, hallucinated references should simply get the submitter banned from ever applying again. Anyone submitting papers or anything with hallucinated references shall be publicly shamed. The problem isn't only the LLMs hallucinating, it's lazy and immoral humans who don't bother to check the output either, wasting everyone's time and corroding public trust in science and research.

I fully agree. Not reading your own references should be grounds for banning, but that's impossible to check. Hallucinated references cannot be read, so by definition,they should get people banned.

Re: Prism

#412

Post by egorfine »

> Chat with GPT‑5.2

> Draft and revise papers with the full document as context

> ...

And pay the finder's fee on every discovery worth pursuing.

Yeah, immediately fuck that.

Re: Prism

#413

Post by lionkor »

arnejenssen wrote: post #374

This assumes that the article, the artifact, is most valuable. But often it is the process of writing the article that has the most value. Prism can be a nice tool for increasing output. But the second order consequence could be that the skill of deep thinking and writing will atrophy. "There is no value added without sweating"

Work is value and produces sweat, and OpenAI sells just the sweat.

Re: Prism

#414

Post by jll29 »

spacebuffer wrote: post #219
efficax wrote: post #44

Earlier quoted context omitted.

Overleaf is a little curious to me. What's the point? Just install LaTeX. Claude is very good at manipulating LaTeX documents and I've found it effective at fixing up layouts for me.

I'd use git in this case, I am sure there are other reasons to use overleaf otherwise it wouldn't exist but this seems like a solved issue with git.

You can use actually git (it's also integrated in Overleaf).

You can even export ZIP files if you like (for any cloud service, it's not a bad idea to clone your repo once in a while to avoid begin stuck in case of unlikely downtime).

I have both a hosted instance (thanks to Overleaf/ShareLaTeX Ltd.) and I'm also paying user for the pro group license (>500€/year) for my research team. It's great - esp. for smaller research teams - to have the maintenance outsourced to a commercial provider.

On a good day, I'd spend 40% in Overleaf, 10% in Sublime/Emacs, 20% in Email and 10% in Google Scholar/Semantics Scholar and 10% in EasyChair/OpenReview, the rest in meetings.

Re: Prism

#415

Post by jll29 »

BlueTemplar wrote: post #281
vicapow wrote: post #35

Earlier quoted context omitted.

The deeper I got, the more I realized really supporting the entire LaTeX toolchain in WASM would mean simulating an entire linux distribution :( We wanted to support Beamer, LuaLaTeX, mobile (wasn't working with WASM because of resource limits), etc.

But what's the point ? To end up with yet another shitty (because running inside a browser, in particular its interface) web app ? Why not focus efforts into making a proper program (you know, with IBM menu bars and keyboard shortcuts), but with collaborative tools too ?

You are right in pointing out that the Web browser isn't the most suitable UI paradigm for highly interactive applications like a scientific typesetting system/text editor.

I have occasionally lost a paragraph just by accidental marking a few lines and pressing [Backspace].

But at the moment, there is no better option than Overleaf, and while I encourage you to write what you propose if you can, Overleaf will be the bar that any such system needs to be compared against.

Re: Prism

#416

Post by sn0wr8ven »

mfld wrote: post #408

I'd like to hypothesize a little bit about the strategy of OpenAI. Obviously, it is nice for academic users that there is a new option for collaborative LaTeX editing plus LLM integration for free. At the same time, I don't think there is much added revenue expected here, for example, from Pro features or additional LLM usage plans. My theory is that the value lies in the training data received from highly skilled ac…

It is nice for academics, but I would ask why? These aren't tasks you can't do yourself. Yes it's all in one place, but it's not like doing the exact same thing previously was ridiculous to setup.

A comparison comes to mind is the n8n workflow type product they put out before. N8n takes setup. Proofreading, asking for more relevant papers, converting pictures to latex code, etc doesn't take any setup. People do this with or without this tool almost identically.

Re: Prism

#417

Post by egorfine »

JBorrow wrote: post #32

From my perspective as a journal editor and a reviewer these kinds of tools cause many more problems than they actually solve. They make the 'barrier to entry' for submitting vibed semi-plausible journal articles much lower, which I understand some may see as a benefit. The drawback is that scientific editors and reviewers provide those services for free, as a community benefit. One example was a submission their und…

> these kinds of tools cause many more problems than they actually solve

For whom? For OpenAI these tools are definitely the solutions. They are developing by throwing various AI-powered stuff at the wall to see what sticks. These tools also demonstrate to the investors that innovation did not stall and to show that AI usage is growing.

Same with Microsoft: none of the AI stuff they are shoving down the users' throats were actually designed for the users. All this stuff is only for the token usage to grow for the shareholders to see.

Similar with Google although no one can deny real innovation happening there.

Re: Prism

#418

Post by 3form »

karmakurtisaani wrote: post #373
Phelinofist wrote: post #363

Earlier quoted context omitted.

FWIW, my immediate reaction was the same "That reminds me of NSA PRISM"

Same here.

Same, to the point where I was wondering if someone deliberately named it so. But I expect that whoever made this decision simply doesn't know or care.

Re: Prism

#419

Post by jll29 »

jmdaly wrote: post #283
jasonfarnon wrote: post #224

Earlier quoted context omitted.

don't think so. I think latex was one of academics' earlier use cases of chatgpt, back in 2023. That's when I started noticing tables in every submitted paper looking way more sophisticated than they ever did. (The other early use case of course being grammar/spelling. Overnight everyone got fluent and typos disappeared.)

It's funny, I was reading a bunch of recent papers not long ago (I haven't been in academia in over a decade) and I was really impressed with the quality of the writing in most of them. I guess in some cases LLMs are the reason for that!

I recently got wrongly accused of using LLMs to help write an article by a reviewer. He complained that our (my and my co-worker's) use of "to foster" read "like it was created by ChatGPT". (If our paper was fluent/eloquent, that's perhaps because having an M.A. in Eng. lit. helped for that.)

I don't think any particular word alone can be used as an indicator for LLM use, although certain formatting cues are good signals (dashes, smileys, response structure).

We were offended, but kept quiet to get the article accepted, and we changed some instances of some words to appease them (which thankfully worked). But the wrong accusation left a bit of a bad aftertaste...

Re: Prism

#420

Post by andy12_ »

biophysboy wrote: post #289
direwolf20 wrote: post #263

Earlier quoted context omitted.

Translation is something Large Language Models are inherently pretty good at, without controversy, even though the output still should be independently verified. It's a language task and they are language models.

Are they good at translating scientific jargon specific to a niche within a field? I have no doubt LLMs are excellent at translating well-trodden patterns; I'm a bit suspicious otherwise..

In my experience of using it to translate ML work between English->Spanish|Galician, it seems to literally translate jargon too eagerly, to the point that I have to tell it to maintain specific terms in English to avoid it sounding too weird (for most modern ML jargon there really isn't a Spanish translation).
Post reply on HN