Top
Best
New

Posted by meetpateltech 1/27/2026

Prism(openai.com)
781 points | 524 commentspage 8
legitster 1/27/2026|
It's interesting how quickly the quest for the "Everything AI" has shifted. It's much more efficient to build use-case specific LLMs that can solve a limited set of problems much more deeply than one that tries to do everything well.

I've noticed this already with Claude. Claude is so good at code and technical questions... but frankly it's unimpressive at nearly anything else I have asked it to do. Anthropic would probably be better off putting all of their eggs in that one basket that they are good at.

All the more reason that the quest for AGI is a pipe dream. The future is going to be very divergent AI/LLM applications - each marketed and developed around a specific target audience, and priced respectively according to value.

Otterly99 1/28/2026||
I completely agree.

In my lab, we have been struggling with automated image segmentation for years. 3 years ago, I started learning ML and the task is pretty standard, so there are a lot of solution.

In 3 months, I managed to get a working solution, which only took a lot of sweat annotating images first.

I think this is where tools like OpenCode really shine, because they unlock the potential for any user to generate a solution to their specific problem.

falcor84 1/27/2026||
I don't get this argument. Our nervous system is also heterogenous, why wouldn't AGI be based on an "executive functions" AI that manages per-function AIs?
noahbp 1/27/2026||
They seem to have copied Cursor in hijacking ⌘Y shortcut for "Yes" instead of Undo.
drusepth 1/28/2026|
In what applications is ⌘Y Undo and not ⌘Z? Is ⌘Y just a redundant alternative?
zerocrates 1/28/2026||
Ctrl-Y is typically Redo, not Undo. Maybe that's what they meant.

Apparently on Macs it's usually Command-Shift-Z?

camillomiller 1/28/2026||
Given what Prism was at the NSA, why the hell would any tech company greenlight this name?
tzahifadida 1/28/2026||
Since it offers collaboration for free, it can take a bite out of overleaf market.
ai_critic 1/27/2026||
Anybody else notice that half the video was just finding papers to decorate the bibliography with? Not like "find me more papers I should read and consider", but "find papers that are relevant that I should cite--okay, just add those".

This is all pageantry.

sfink 1/27/2026||
Yes. That part of the video was straight-up "here's how to automate academic fraud". Those papers could just as easily negate one of your assumptions. What even is research if it's not using cited works?

"I know nothing but had an idea and did some work. I have no clue whether this question has been explored or settled one way or another. But here's my new paper claiming to be an incremental improvement on... whatever the previous state of understanding was. I wouldn't know, I haven't read up on it yet. Too many papers to write."

renyicircle 1/27/2026|||
It's as if it's marketed to the students who have been using ChatGPT for the last few years to pass courses and now need to throw together a bachelor's thesis. Bibliography and proper citation requirements are a pain.
pfisherman 1/27/2026|||
That is such a bummer. At the time, it was annoying and I groused and grumbled about it; but in hindsight my reviewers pointed me toward some good articles, and I am better for having read them.
olivia-banks 1/27/2026|||
I agree with this. This problem is only going to get worse once these people enter academia and facing needing to publish.
olivia-banks 1/27/2026|||
I've noticed this pattern, and it really drives me nuts. You should really be doing a comprehensive literature review before starting any sort of review or research paper.

We removed the authorship of a a former co-author on a paper I'm on because his workflow was essentially this--with AI generated text--and a not-insignificant amount of straight-up plagiarism.

NewsaHackO 1/27/2026||
There is definitely a difference between how senior researchers and students go about making publications. To students, they get told basically what topic they should write a paper on or prepare data for, so they work backwards: try to write the paper (possibly some researching information to write the paper), then add references because they know they have to. For the actual researchers, it would be a complete waste of time/funding to start a project on a question that has already been answered before (and something that the grant reviewers are going to know has already been explored before), so in order to not waste their own time, they have to do what you said and actually conduct a comprehensive literature review before even starting the work.
black_puppydog 1/27/2026|||
Plus, this practice (just inserting AI-proposed citations/sources) is what has recently been the front-runner of some very embarrassing "editing" mistakes, notably in reports from public institutions. Now OpenAI lets us do pageantry even faster! <3
verdverm 1/27/2026|||
It's all performance over practice at this point. Look to the current US administration as the barometer by which many are measuring their public perceptions
teaearlgraycold 1/27/2026|||
The hand-drawn diagram to LaTeX is a little embarrassing. If you load up Prism and create your first blank project you can see the image. It looks like it's actually a LaTeX rendering of a diagram rendered with a hand-dawn style and then overlayed on a very clean image of a napkin. So you've proven that you can go from a rasterized LaTeX diagram back to equivalent LaTeX code. Interesting but probably will not hold up when it meets real world use cases.
adverbly 1/27/2026|||
I chuckled at that part too!

Didn't even open a single one of the papers to look at them! Just said that one is not relevant without even opening it.

maxkfranz 1/27/2026|||
A more apt example would have been to show finding a particular paper you want to cite, but you don’t want to be bothered searching your reference manager or Google Scholar.

E.g. “cite that paper from John Doe on lorem ipsum, but make sure it’s the 2022 update article that I cited in one of my other recent articles, not the original article”

thesuitonym 1/27/2026||
You may notice that this is the way writing papers works in undergraduate courses. It's just another in a long line of examples of MBA tech bros gleaning an extremely surface-level understanding of a topic, then decided they're experts.
r_thambapillai 1/28/2026||
didn't OpenAI just say they needed a code red to be relentlessly focussed on making ChatGPT market leading again? Why are they launching new products? Is the code red over is the gemini threat considered done?
unixzii 1/28/2026||
It may be useful, but it also encourages people to stop writing their own papers.
mves 1/28/2026|
As they demo in the video, it even encourages people to actually skip doing the research (which includes reading both relevant AND not-so-relevant papers in order to explore!) Instead, prompt "cite some relevant papers, please", and done. Hours of actual reading, thinking, and exploration reduced to a minimum.

A couple of generations of students later, and these will be rare skills: information finding, actual thinking, and conveying complex information in writing.

0dayman 1/27/2026||
in the end we're going to end up with papers written by AI, proofread by AI .....summarized for readers by AI. I think this is just for them to remain relevant and be seen as still pushing something out
falcor84 1/27/2026|
You're assuming a world where humans are still needed to read the papers. I'm more worried about a future world where AIs do all of the work of progressing science and humans just become bystanders.
drusepth 1/28/2026||
Why are you worried about that world? Is it because you expect science to progress too fast, or too slow?
falcor84 1/28/2026||
Too fast. It's already coding too fast for us to follow, and from what I hear, it's doing incredible work in drug discovery. I don't see any barrier to it getting faster and faster, and with proper testing and tooling, getting more and more reliable, until the role that humans play in scientific advancement becomes at best akin to that of managers of sports teams.
chaosprint 1/27/2026||
As a researcher who has to use LaTeX, I used to use Overleaf, but lately I've been configuring it locally in VS Code. The configuration process on Mac is very simple. Considering there are so many free LLMs available now, I still won't subscribe to ChatGPT.
delduca 1/27/2026|
First 5 seconds reading and I have spotted that was written by AI.
drusepth 1/28/2026|
We human writers love emdashes also ;)
More comments...