[josefritzishere]

I tried using AI last week for some lazy research. The results were highly questionable. I asked it to cite sources. It did but those sources were even more questionable. AI has no concept of objective reality. Sometimes it is glaringlyobvious that it's just scraping 4chan and facebook and barfing up trash.

[uberman]

I was of the same mindset and did a similar test that also failed spectacularly. I asked for authoritative pros and cons on a subject in my field and asked it to cite sources. The pros and cons were ok, but the sources completely made up.

So, not so great there.

Then recently I had to modernize a python app written by someone who no longer works for the organization and was circa python 3.6. Several of the key libraries no longer supported the interfaces being called and there was no documentation at all.

On a whim I asked an LLM to help modernize the code, file by file and it cut the effort in half.

So, pretty great there.

[mattgreenrocks]

LLMs are really good for things that don’t have to be 100% correct, presuming that the artifact generated is checked by someone with expertise. We used it to suggest some places to visit in Italy the past year. Great use case for AI: it brainstorms spots for an itinerary that we then check for ourselves to see if it fits our requirements. The machine helps us think beyond the usual destinations, and we vet the results. Hallucinations are not an issue.

Hallucinations seem fundamental to how LLMs work now, so AI is probably still a ways off from being able to absorb responsibility for decisions like a human can.

I’m sure this comment will offend both pro and anti AI camps. :)

[coro_1]

We found AI is surprisingly good at programming (when no one thought this would be the frontier) which says a lot about the specific nature of the tools.

You can for example, do minimal input (some peppered phrases) and see a rich answer. More specificity and research on your own, brings it out to play. The designs appear to level their self to your own capabilities.

[actuallyrizzn]

i hate being a broken record on this, but skill issue.

if you're a senior researcher that understands how to research, and you send a baby junior researcher off to do the work and did not give them tips or parameters to improve the research process, that's on you, not the junior.

[simonw]

Which tool and model?