News for hackers. Stuff that matters.
Re:Wikipedia-grounded chatbot “outperforms all baselines” on... by oxguy3 on Monday July 17, 2023 @03:10PM (#36762618)
The Wikipedia community has generally been pretty resistant to allowing fully AI-based tools in. We've had tools such as Lsjbot (https://en.wikipedia.org/wiki/Lsjbot) in the past, but they've failed to gain community consent on any of the large Wikipedias. If someone tries to bring an LLM-based tool to Wikipedia, it would take a lot of finesse to have any shot of the community allowing it.
Re:Wikipedia-grounded chatbot “outperforms all baselines” on... by bruce343434 on Monday July 17, 2023 @03:41PM (#36763005)
I don't think it takes much finesse to just randomly start "improving" articles using the output of an LLM. It only takes a single well meaning yet misguided person. Remember this? https://www.theguardian.com/uk-news/2020/aug/26/shock-an-aw-...
Re:Wikipedia-grounded chatbot “outperforms all baselines” on... by oxguy3 on Monday July 17, 2023 @07:50PM (#36765784)
Yeah, definitely a potential problem on the smaller language Wikipedias as the Scots Wikipedia incident shows, but for the big ones, low-quality content from new editors is not really a new problem to deal with.
Re:Wikipedia-grounded chatbot “outperforms all baselines” on... by 1270018080 on Monday July 17, 2023 @05:18PM (#36764255)
But what about the whole mass of tech bros who don't understand what LLM's are (random text generators and nothing more), and manually start to add changes? It's a virus polluting every industry.