LesdmarkDec 6, 2023
Elio Campitelli

There's not enough "fuck you"s in the world to react to this shit. #LLMs should be tools used in the service of people; what in the world is this proposal to make people work for LLMs?!

Any and all changes to scientific publishing needs to be for so that other **people** can access them and understand them.

And the single most important change would be for Nature and other publishers not to charge 29.99 USD for a shitty 4-paragraph essay that they didn't pay for themselves.

#AcademicChatter

Dec 5, 2023 at 4:38pmPhanpy
Show threadNeurocitizenDec 5, 2023
@eliocamp jajaja lo que faltaba
Show threadScott CainDec 5, 2023
@eliocamp While I'm certainly no fan of Nature, I'm sympathetic to the idea that making papers easier to parse would be a good thing. I work with many curators of biological papers at various model organism databases (whose goal is to synthesize data for researchers to facilitate what they do). Improving their throughput would be a net positive for the research communities they serve.
Show threadKate Nyhan is changing serversDec 5, 2023

@scottcain @eliocamp
Sure, but you'd make papers easier to parse by
- having the authors comply with reporting guidelines
- having the publishers provide correctly tagged JATS XML
- disseminating the paper with a clearly stated license to explicitly allow reuse
- offering a free and well-documented API

Writing in the third person and using APA instead of Vancouver, hoping that that will make spicy autocorrect more accurate? I don't think that's the future of synthesizing scientific knowledge.

Show threadSharon MachlisDec 5, 2023
@eliocamp "You should fundamentally change the way you write scientific papers *so that people don't have to bother reading your paper* and instead can ask an LLM to summarize it" is an interesting take.
Show threadSchroedingers_CatDec 6, 2023
@smach @eliocamp "Write papers so LLMs can summarise them then I can reference them without reading the original" is not my idea of progress.
Show threadGoodNewsGreyShoes🔞(on hiatus)Dec 8, 2023
@SteveClough @smach @eliocamp ☝️👍😵‍💫
Show threadPete Alex Harris🦡🕸️🌲/∞🪐∫Dec 6, 2023

@smach @eliocamp

We already have a summary of every scientific paper, it's called the abstract.

There doesn't seem to be a *use* for an unreliable automatically generated summary that you can't cite.

Show threadDavid Weir Dec 5, 2023

@eliocamp This letter has to be a joke, right?

If LLMs can’t be made to do what their controllers want by ingesting abstracts and other pre-existing, highly structured, bibliographic data, the solutions are:

1) fix the LLM
or better 2) just don’t bother using an LLM.

Show threadTimSalabimDec 5, 2023
@eliocamp how bout training a LLM to write that style of paper for us by providing the human written paper, the code and the data?
But then, large parts of the code were probably written by a LLM in the first place... AI cannibalism at its best.
Show threadjoachimDec 5, 2023
@eliocamp If LLMs are so clever (sarcasm!), why can't they understand the format of a cited reference? After all, it's really just a special format of hyperlink!
Show threadDavid O'BrienDec 5, 2023

Indeed.

Also, this simply adds more evidence to the growing pile that #LLMs are very far from "intelligent", whatever that means.

@eliocamp

Show threadDarryl RammDec 5, 2023
@eliocamp in the nicest way possible could Springer Nature just fuck off and die.
Show threadElisa Fadda-🔺Dec 5, 2023
@eliocamp like we need more time wasting on formatting.... so a machine can read them????... that sounds sensible... 🤔🤔 or maybe the editorial board at Nature has been already swapped with ChatGPT 🤣
Show threadRené Vincent JansenDec 5, 2023
@eliocamp @pluralistic That reminds me of the sixties/seventies when we were told that fonts needed to be adapted to fit cathode screens and needed to be blocky to be machine readable.
Show threadJonDec 6, 2023
@rvjansen @eliocamp @pluralistic how many distinguishable font families can be had from a 5x7 dot matrix? I was just happy when we had both upper and lower case.
Show threadRené Vincent JansenDec 6, 2023
@oddhack @eliocamp @pluralistic I was thinking of https://en.m.wikipedia.org/wiki/New_Alphabet ; yes I also remember the first lower case printer train we got!
New Alphabet - Wikipedia

Show threadJonDec 6, 2023

@rvjansen @eliocamp @pluralistic oh, you mean vector displays! Makes sense though if you consider serifs and such vs. the limited drawing speed, simpler glyphs probably were a sensible compromise to get more text out.

My exposure to those was largely limited to the PS300 and the Atari vector hardware used for Tempest / Gravatar / etc.

Show threadMichael SchneiderDec 5, 2023
@eliocamp "Describing the research method rather than citing a reference to it" would be something that mostly benefits any reader not too familiar with the research domain. I would have liked this very much while I was still in university, and I think older papers used to do this.
Show threadJoris MeysDec 5, 2023
@eliocamp they're taking the piss, right? Right??
Show threadTDec 5, 2023
@eliocamp LOL, now LLMO will be the new SEO.
Show threadOliver SampsonDec 5, 2023
@eliocamp Whoa. Wait. What is this fresh hell?
Show threadLee Fife - old accountDec 5, 2023
@eliocamp I'll ... ah .... get right on that ...
Show threadAntonio Páez 🇲🇽🇨🇦Dec 5, 2023
@eliocamp @enjoyingkink
Show threadStewart SimsDec 5, 2023
@eliocamp modifying your training data to try and get your AI to perform better is a great way to build useless AI.
Show threadJPDec 5, 2023
@eliocamp shades of "autonomous vehicles will work great, we promise, if every human goes around wearing bluetooth tags"
Show threadmccDec 6, 2023
@eliocamp On the other hand pondering a world where academic papers can submit a separate file/field for the AI summary, and everyone writing a paper starts filling this space with 4K of random binary data including the EICAR virus string, malformed UTF-8, garbage data and libelous statements about past and current U.S. Presidents
Show threadBlake C. StaceyDec 6, 2023
@eliocamp Every sentence in that calls out for a "No. No. Fuck you. No. Fuck you sideways."
Show threadBlake C. StaceyDec 6, 2023
@eliocamp Something tells me that Nature would not look kindly upon a letter that reads, "Counterpoint: Take Sam Altman's cock out of your mouth before you start talking".
Show threadDustin D. WindDec 6, 2023

@eliocamp My cynically-optimist self wants to think maybe the author was trying to help us know how to screw them up. 😈

As a coder friend let me know, if you're opposed to knowingly being mined for AI data then be aware that "no data" does indeed screw things up. But not nearly as much as "bad data" does. Remember, it took trolls no time at all to turn a famous chatbot attempt into a mental case spewing racist profanities.

Show threadLuka RubinjoniDec 6, 2023
@eliocamp I wish Nature would print my shitty 4-paragraph essays...
Show threadAndrew HyattDec 6, 2023
@eliocamp papers being in PDF certainly is non optimal for reading, particularly on mobile. Using an easier format for both LLMs and my browser (such as HTML) would be a welcome change.
Show thread⁂ Fish Id WardrobeDec 6, 2023
@eliocamp "Papers shouldn't cite references because that's an obvious way we can tell LLMs get it wrong"? Nice.
Show threadKonrad HinsenDec 7, 2023
@eliocamp Alternative proposal: Nature should publish *only* summaries, written by the authors of the paper, referring to the paper itself as published in a non-profit Open Access journal with good peer reviewing practices.
Show threadJ.H.NoyesDec 8, 2023

@eliocamp

Anyone remember that ridiculous paper that said low gut serotonin was the cause of Long Covid? That's the type of convoluted, low quality work one should expect from over-reliance on LLMs in the research space. If we aren't careful, LLMs will usher us into a new dark age where it becomes impossible to separate truth from fiction.

@accretionist

Show threadAearil  Dec 8, 2023
@eliocamp Beyond the ridiculousness of the idea itself, the worst about this is the fact they're asking for transformations that are already possible to automate using good ol' programming --'
Show threadZachary RosenDec 9, 2023
@eliocamp I won’t censor my words on this one. Whatever tech bro wrote this particular correspondence can get fucked. Academic register sucks already for how hard it is to be adequately expressive. Neutering it more just to help someone else’s chatbot make up more misinformation isn’t just ridiculous it’s legitimately insulting.
Show threadElizabeth SudduthDec 30, 2023
@eliocamp What a strange opinion for Nature to publish. Looking at Google Scholar, the author seems to be a geographer whose publication history is mostly in the scholarship of teaching and learning field. I'm not sure why the author submitted this letter to Nature or why it was published
https://scholar.google.co.uk/scholar?as_q=&btnG=Search+Scholar&as_sauthors=%22Terence%2BDay%22
Google Scholar

Show threadSteve HerseyDec 30, 2023
@eliocamp
I heartily recommend that all researchers format their papers to make it impossible for AI tools to process them. At least at the present time, AI is being used in negative ways, and catering to its use is against your best interests.
Show threadThomas Favre-Bulle 🏳‍🌈Dec 30, 2023
@eliocamp @nyrath Describing the research method in the paper itself is human friendly though.