9 comments

[ 3.2 ms ] story [ 38.5 ms ] thread
Bard lies a lot, probably why.

I try this query every once in a while:

Please provide a list of small SUVs sold in 2023, sorted by rear hip width.

It gives a list, but makes up a bunch of numbers for hip width.

I haven't found Bard to be particularly useful due to the lies and hallucinations. I feel like there should be more warning that the information you receive is very likely to be incorrect, given that there's a large number of people who google a question and accept the first answer they find as fact, and they'll be transitioning to this technology, perhaps initially unaware of its shortcomings.
I haven't had a lot of luck with it. With large (tens of MB) PDFs it silently fails to add the source. With smaller PDFs I don't need a machine to summarize them. For dense, factual materials it just invents things that are not true. As an example of the latter, I added a source from the inventory of housing construction opportunity sites in the city of Berkeley, asked some basic factual questions, and it just blends numbers from the source together into a soup of lies.
Seem to be a bit opinionated on how you should use it. But maybe when by the time it becomes more widely available (seem to be restricted to US only currently) it could become a more general tool.
(comment deleted)
It seems to work great to ask questions about any books, for example I was unable to remember the name of a character and just uploading the PDF and asking gave me the correct answer as opposed to the hallucinated response of ChatGPT or Bard.
Although the article says that "Google released NotebookLM to the wider public", availability is still restricted to the USA.