Fake Paper Generator | Hacker News Reader

Fake Paper Generator

fakepaper.app

46 pointsnoah3225 comments

It’s interesting that the fake paper generators of today are lower quality than the ones from over an decade ago. SciGen makes more believable papers and it came out in 2005 [0]

This generator reeks of ChatGPT, and says “whimsical” and “fictional” every other word if you ask it to generate a paper that is obviously untrue. All this “safety” is so boring.

[0] https://pdos.csail.mit.edu/archive/scigen/

Looks like maybe it doesn’t work anymore which is a little sad.

i mean its certainly designed to be funny not realistic or believable. and yes it uses openai apis for content generation.

It's not at all funny. It's dumb, and from the headline alone - "A dairy delight" - I already knew it was the lame and predictable ChatGPT language model with all it's repetitiveness, puns and consciously designed idiocy.

"dairy-inspired exploration" - word salad garbage

"we will humorously consider" - good to be reminded again, I was already falling for it

"the implications of a lactose-laden lunar surface" with acutely annoying aliterations.

it started as a joke about boeing doors falling off and was quite funny. unfortunately i don't have time write a paper for each person who clicks the generate button so this will have to do.

To clarify, my comment was directed at chatGPT itself, not your project.

It was funny. It will be funny again. It just wasn’t -that- funny - this- time.

Sure, I'm a bit tired of repetitive ChatGPT output too. But I'm really surprised at how vitriolic this reply is.

It's a student's side project, with the tagline "Your Source for Academic Satire", with a basic one-liner prompt textbox. No one's expecting this to disrupt the academic world or whatever. The author (a college student) doesn't advertise this to be "the world's most advanced academic scientific rigorous paper generator", of course it's going to hook into GPT for content.

You don't have to like the project or think it's interesting, but problems with ChatGPT output are already apparent and well-known so I'm not sure what your comment adds.

I'm angry at ChatGPT and the like polluting the web, academia and people's minds.

To be fair, I would totally have the same reaction if this project didn't have all the "satire" and "jokes" stamped on it, and if it was advertised for actual use in research papers. And while this project in particular is (relatively) harmless, it is pretty disgusting to see in real paper submissions.

You're not wrong, but I don't know if we should have been celebrating it if the quality of fake paper generators had been improved a lot since 2005.

SCIgen papers are the results of lots of manual rules:

https://github.com/strib/scigen/blob/master/scirules.in

It's an example of https://en.wikipedia.org/wiki/Expert_system

I am not sure what drives fakepaper.app but the output didn't seem to have as much variety.

i can open source it if people want but its llms under the hood

I think the prompt should be changed to commit to the joke a bit more. It keeps reminding me that it's ridiculous satire.

See the sibling comment https://news.ycombinator.com/item?id=41094612

Apparently this is a safety feature.

I think that’s just how ChatGPT works. It’s really tiring and apparently the only way to tell you aren’t talking to a bot in 2030 will be to say you think eating soap is a fantastic idea or something else AI refuses to say because of “safety”.

Claude is much better at these tasks. Look at WebSim for an example.

I think with the right prompting you can easily get subtly ridiculous satire instead of “whimsical”:

*Abstract:*

The hypothesis that the Moon is composed of cheese, though traditionally viewed as a whimsical notion, warrants a serious re-examination through a multifaceted approach. This paper presents compelling evidence from historical texts, optical illusions, and chemical analogies to support this hypothesis. Historical documents reveal that ancient cultures, including the Greeks and medieval Europeans, described the Moon's surface as cheese-like, noting its porous and cratered appearance. Early telescopic observations by Galileo and others recorded surface features that strikingly resemble the texture of certain aged cheeses, such as Emmental and Roquefort.

Furthermore, modern spectroscopic analyses have identified organic compounds in lunar soil samples that share chemical similarities with components found in cheese, including carbonates and amino acids. Notably, the peculiar luminescence of some lunar regions parallels the phosphorescent qualities observed in certain mold-ripened cheeses. Additionally, lunar regolith's high calcium content mirrors the mineral composition of dairy products, suggesting a possible overlooked connection.

By combining historical accounts with contemporary scientific observations and chemical analyses, this paper advocates for a renewed consideration of the Moon's composition. Understanding the intersection of cultural mythology and scientific exploration can offer novel perspectives on celestial bodies and inspire innovative approaches in astrobiological research.

Looks like generations are failing. Didn't plan for 60+ people to be generating concurrently.

attempting to push basic job queue. here goes nothing

looks like it worked

Seeing that the output is just a few pages, I tried generating one that is supposed to be short. It made the title longer and added an entirely different topic:

https://fakepaper.app/storage/f6d24c9c-0285-4144-9f5e-5c5c43...

Compare this with the original:

https://www.ams.org/journals/bull/1966-72-06/S0002-9904-1966...

yeah it was made to do 2-3 pages with a table / other element. i just picked that length as a good length for comedic effect. its llms so like it does things like that

This is the beginning of the end of science.

ooops.

I used to spew out fake papers to my friends aswell, as a joke. The process setup was writing a large prompt that requested gpt to write a title and all the other necessary headlines, then answer each headline with content.

The prompt also asked gpt to output it in LaTeX to which I then fed into Overleaf.

The end result was hilarious scientific papers with looked professional at first sight, until you started reading it. This process allowed me create lots of funny papers in minutes for me and my friends amusement.

Reply on news.ycombinator.com