This is my new favorite Grok stat - ThreadSky | a Reddit-style client for Bluesky

caseynewton.bsky.social • 20 days ago

This is my new favorite Grok stat

A study of eight AI search engines found they provided incorrect citations of news articles in 60%+ of queries; Grok 3 answered 94% of the queries incorrectly (Columbia Journalism Review)

Main Link | Techmeme Permalink

Comments

bawks.com•19 days ago

https://youtu.be/neQwFEzUmiE?si=uFIZGJzyYdsiWEo0

adamkuphaldt.bsky.social•20 days ago

I hope they accounted for the "sass" and "epic, based tone" in that study! Had to account for something.

dailytrix.bsky.social•20 days ago

The only questions it answers correctly are about what a shitshow Elon and DOGE are.

firthermor.bsky.social•20 days ago

So even with this abysmal record, Grok may be still be correct more often than Elon Musk.

mmg1960s.bsky.social•19 days ago

Musk is just like Zuckerberg, if it's their original idea, it sucks.

mikecavanagh.bsky.social•19 days ago

Grok pooched one i used. Gratuitous hallucinations

lucas.meyerperin.org•19 days ago

One of the scariest things I’ve seen recently is a lot of people on X asking “grok is this true?” to any tweet critical of their position. Grok doesn’t do a horrible job, but does a bad enough job to misinform people on average.

lucas.meyerperin.org•19 days ago

Yesterday, during the market meltdown, Grok said that it was likely because of DeepSeek, given that tech companies were suffering the most. Completely ignored Tesla.

maffe.is•19 days ago

Hey, it’s more consistent!

rsquared.bsky.social•20 days ago

The Ivory soap of LLMs

ncxiao.bsky.social•19 days ago

Hi Grok, tell me five things you did last week. Hmm, cited 100 references, cited 100 references, cited 100 references, cited 100 references, and cited 100 references. Actually, only cited 6 in each because the other 94 were all wrong. I can be morewrong if you want.

eiggen.bsky.social•20 days ago

More melting down of glaciers will probably improve things though

kenfromchicago.bsky.social•19 days ago

alajon.bsky.social•20 days ago

but it was wrong while flirty 😏😏😏

tthhoorr.bsky.social•19 days ago

ChatGPT and Gemini too. Especially when using the search agents on these models to produce any kind of quantitative analysis across multiple sources (financial reports, scientific studies, surveys etc) they will constantly invent numbers and hallucinate basic arithmetic. Completely useless “PHDs” 😂

dawnwilliamson.bsky.social•19 days ago

Grok doesn’t like EM. I hope Grok didn’t get that one wrong. 🤣

mitch-henke.bsky.social•20 days ago

I would think one of the most vital things a genAI could learn would be how to tell truth from fiction, and how to correctly cite a source. If they can't do that, it's all just hot air and noise.

quinum.bsky.social•20 days ago

Shit! I hate when that happens.

Comments

Posting Rules

Reply