WelcomeWEFreedomWindsorEssexCounty-ActionCommunicationsRalliesResourcesSocialEventsWednesdayNightMeetings-ActivismAgoraAgorismCanadaCommonLawCommunityDataCentersDataCentresDoocracyFarmersMarketIntroductionsNaturalLawNewslettersOntarioOrganizingOutreachProjexProjexWikiRalliesResistanceAnalysisSoireesSoiréesWikiSpooksWindsorElectionsTop Subs
5
Tech - - (+5|-1)
Can LLMs be used to look things up?(self)
6 commentsbyLarrySwinger

If a professor were to make as many mistakes as ChatGPT, he'd lose his tenure in a heartbeat. Not that tenure means anything because Jordan Peterson lost his but this is beside the point. In spite of their flaws we need them for looking things up because all the search engines are almost completely worthless now. It's weird: in 2010 you were able to formulate your query as a question and the results were more useful than what ChatGPT spits out now. We're condemned to interacting with a stupid bot now.

I realized that if you ask an LLM about a subject you already know something about, it will state some facts which you can verify to be true or not. First of all, this gives you an indication of the reliability of this answer unto itself. If it lists facts which you know to be true, the other facts are more likely to be true as well. It has simply gathered good information for this query.

Secondly, if it uses deductive reasoning to reach its conclusion, you can simply follow along to verify that it's internally consistent. Thus you can learn new things.

How do you guys feel about using LLMs for search queries?

Comment preview

[-]wmjofus5(+4|0)

What is an 'LLM'?

parentlinkreply
[-]Neol3(+2|0)

I think LLMs are good for summarizing something - I often prompt them to summarize debates on some topic or consensus on some topic. Factual stuff (especially measured in numbers) is better to be found on old school websites.

Also they are good with showing keywords and terms which you needed for searching but didn't know those existed before asking.

parentlinkreply
[-]LarrySwinger3(+2|0)

Also an approach that is occuring to me is to simply expect it needs some prompting. Mistral Vibe provides sources (without requiring registration). So you can check the sources to see if they agree and, if not, simply state so and ask it to look for better sources or double-check stuff. I do get better results beyond the first prompt. You can kind have to ignore the confidence in his tone, that's mostly what is offensive to me.

parentlinkreply
[-]RickSanchez3(+2|0)

I think Google has done that with Gemini. When you search on Google, the first result you get is the Gemini AI chat.

parentlinkreply
[-]cyberbully0(+1|-2)

Jordan Peterson is a fucking jew lover

parentlinkreply