#llmfail — Public Fediverse posts

1. Invent an illness;
2. Put up a couple of preprints about it;
3. See AI companies' web crawlers swallow it hook, line, and sinker whole;
4. LLMs start telling users that the condition is real. WTF.
5. See published papers cite them, likely from authors using LLMs to write them. WTAF.

https://www.nature.com/articles/d41586-026-01100-y

#LLMFail #LLMs #chatbots #Bixonimania

35m < 16m
#LLMfail

That's some next level Bond villain shite!

Imagine having an ego that requires that kind of virtual massaging... by what amounts to a pre-programmed bot.

https://www.rollingstone.com/culture/culture-features/elon-musk-grok-lebron-james-1235469906/

#prayforhumanity #itisnotAI #LLM #LLMFAIL #LLMprogrammingFAIL #internet

Können wir bitte aufhören, LLMs zu benutzen? Ich habe gerade ein bisschen mit rufus ai (beta) von amazon gespielt.

Nein, LLMs "verstehen" nichts.

#aiFail #llmFail #llm #ai #ki

Habe dies von einer Bekannten bekommen, und ich liebe alles daran. #LLMfail

Here’s a good example: https://www.fishiology.com/flag-cichlid-mesonauta-festivus/

Fishiology.com would like you to know that their author “Michelle” is “a long-time freshwater fish enthusiast with a passion for sharing knowledge about this fascinating hobby.”

In fact, she’s *so* passionate about it that she’s posted five or more long articles a day about it, 7 days a week, for *years*! Gosh, what passion! Why, one could say it’s almost *inhuman*! 🙄

And yet that’s one of the better sites, in that much of its suggestions and info here seem pretty accurate… but the point is that it’s entirely impossible to be sure unless you already know a fair amount on the subject.

In contrast, forum posts may be from people with a wide range of experiences and knowledge, but at least they’re *people* trying to provide genuine answers/anecdotes, usually in good faith.

#aquarium #aislop #llmfail

The aquarium/fishkeeping web has been absolutely *destroyed* by LLMs. It was full of SEO spam previously, but now it’s practically impossible to find semi-trustworthy information.

Example: I poked around for info about tankmates that are reasonably safe to keep alongside South American cichlids such as flag cichlids. Most results were obviously LLM-generated given the recognizable writing style.

The danger is that these pages were *mostly* correct, but also contained a lot of dubious info. LLMs are doing the exact opposite of what they purport to do: by displacing more reliable info with less reliable info that’s conveyed with faux confidence, they’re making the web essentially useless as an information source.

(Which is quite a feat, considering it was full of dubious info to begin with!)

#aquarium #aislop #llmfail

My daughter just came up with a great exercise: challenge your students to find the title of your PhD using ONLY LLMs (no Google allowed). If any of them manage, they get gummy bears 😃

I asked five different models, and got five different answers, all five of which were completely wrong 😂

#AI #ChatGPT #AISlop #LLM #LLMFail #Education #HigherEducation #AcademicChatter

🤖 Think your AI assistant can really reason? Apple’s puzzle tests say otherwise.
📉 See how “thinking” AIs collapse when logic gets real — and why we might be projecting intelligence where there is none.

Hashtags:
#AIReasoning #ChainOfThought #LLMFail #DeepTech

URL:
https://medium.com/@rogt.x1997/the-illusion-of-thought-why-reasoning-ai-might-be-smarter-than-us-but-not-wiser-73427af99baa

I’m sorry, but I cannot help a tiny bit of Schadenfreude. A colleague is an enthusiastic user of ChatGPT and recently told me that one does not need traditional reference managers like Zotero anymore, since you can just ask the LLM to re-format your references according to a given style. Now he got article proofs back with countless comments that the dates in in-text references don’t match the dates in the bibliography. 🙃 #LLMfail

I really love shitty automation. Just posted a rant about ai/llm and tagged as such; not a second later and I get 3 fucking boosts of my introduction post.

It's just so funny; not even that can it handle correctly.

#aifail #llmfail

@peter_mcmahan If some researchers and evaluators can't be bothered to write the questions, analyze the data, and report findings

- and all of these are things I've heard researchers publicly crowing in pride about -

then why should people bother responding?

#evaluation #research #LLMfail #RealEvalTalk

I absolutely love this backfiring use of GPT. I wonder how many customer service reps were let go as part of the business case for this one? #fail #DPDFail #GPTFail #llmfail https://www.theguardian.com/technology/2024/jan/20/dpd-ai-chatbot-swears-calls-itself-useless-and-criticises-firm

I was upset and troubled to see a large language model interpretation of a well known political image, suggesting that wealth hoarders in society foment differences and prejudice between those with little, to protect their wealth.

The LLM summary returned:
"a painting of two men sitting at a table with a plate of food"

what the {expletive deleted}?!?
There are Three People!
How can any interpretation be that there are two people in this image?!?

#llm #llmfail #disgusted #unacceptable

Chatbot-generierte Pilzführer bei Amazon hatte ich bisher nicht auf dem Schirm beim Thema „tödliche Gefahren durch künstliche Intelligenz“.
https://www.techtimes.com/articles/295863/20230901/beware-ai-written-mushroom-foraging-guides-amazon-experts-warn.htm
#ai #llmfail #llm #pilze

@ashley 🙀

Edited:

I was really hoping for an #LLMFail here, but the source site has both definitions.

So, old school extractive summarization fail.

Still, hilarious!