Likewise, "this tool which makes researchers and officer workers 5-10% more efficient" is, like, a big deal, but does not justify the current market valuation of Nvidia, so instead we have to keep trying to invent god or the market implodes.
Replies
Well, and you've got companies bragging about this (gift link) "Since May, BenchSci has cut 23 per cent of staff – about 83 jobs – as it goes all-in on adopting generative AI to do work formerly done by humans" It's doing contract review, that'll be fun ... www.theglobeandmail.com/gift/03017b5...
aug 13, 2025, 10:43 pm • 3 0 • view
guess we'll find out?
aug 13, 2025, 10:43 pm • 1 0 • view
“Format a bibliography from the footnotes in this paper” is a great use for it, honestly.
aug 13, 2025, 10:41 pm • 35 1 • view
From what I hear, "make this a CSV" works surprisingly well, but I do wonder how much little fixes someone telling me "surprisingly well" papers over.
aug 14, 2025, 3:55 pm • 2 0 • view
The standard has to be "does it save time" without introducing horrible errors - if you have a source document, checking the output is there.
aug 14, 2025, 3:58 pm • 2 0 • view
It’s one of the things I use it for. Or to find anomalies in large csv datasets. It’s also decently good at making heat maps. (All the above with a human in the loop, of course).
aug 14, 2025, 4:01 pm • 1 0 • view
Yes, I know it can be done in Python/R/whatever, but I am not allowed to install those on company computers.
aug 14, 2025, 4:03 pm • 1 0 • view
It’s great with standard forms. I’ve used it to create PRD’s (product requirement document) Invoices and such
aug 14, 2025, 4:16 pm • 1 0 • view
“Vectorize this code that I wrote” or “help me with this regex” are also good.
aug 13, 2025, 10:42 pm • 19 0 • view
now you have two problems
aug 13, 2025, 10:53 pm • 17 0 • view
Again - automate improving something you already understand and know that it works.
aug 13, 2025, 11:27 pm • 8 0 • view
LLMs can be good if it is easy to verify the correctness of their output
aug 14, 2025, 12:17 am • 10 0 • view
I think that’s where having domain knowledge already helps making LLMs helpful - you can catch clear challenges and nudge it along paths from your experience. Tbf I wish people thought critically about internet posts …. Generally. Overly trusting the internet is a bad idea.
aug 14, 2025, 1:01 am • 7 0 • view
westlaw has a plugin now for ms-word that automatically generates a table of authorities for a brief and honestly it works great
aug 13, 2025, 11:18 pm • 3 0 • view
Zotero.
aug 13, 2025, 11:43 pm • 2 0 • view
Many, many of my students report that trying this results in hallucinated entries that they have to go through carefully and prune out & it’s easier to just do it themselves
aug 13, 2025, 11:08 pm • 7 0 • view
I’ve already copied out the footnotes here. Hallucinations are much less frequent with set sources.
aug 13, 2025, 11:28 pm • 0 0 • view
You’re not wrong, but I watch LLMs fuck this up all the time.
aug 13, 2025, 10:51 pm • 7 0 • view
Oh I mean, check it. It’s just soulless and mindless and I’ll take the speed up.
aug 13, 2025, 11:24 pm • 1 0 • view
as much as I bash AI, did that once extracting a bibliography from a paper with footnotes and I admit I was impressed how well it did, just a few tweaks needed
aug 13, 2025, 11:27 pm • 3 0 • view
I’m doing this for personal note-taking, not for something I’m handing in.
aug 13, 2025, 11:29 pm • 4 0 • view
I take notes like I’m doing HVI hunting. Bibliographies are great for this.
aug 13, 2025, 11:31 pm • 4 0 • view
It never seems to speed me up…but that’s probably on me.
aug 14, 2025, 1:26 am • 1 0 • view
Honestly? Zotero already does this and the only fuckups or hallucinations that it generates are mine and mine alone.
aug 13, 2025, 11:42 pm • 1 0 • view
I have Zotero, its free!
aug 14, 2025, 12:22 am • 1 0 • view
AI can even make workers less effective. A study found that specialists got worse at detecting cancer by themselves after they had used AI for awhile. Apparently their skills decayed rather quickly. time.com/7309274/ai-l...
aug 14, 2025, 3:54 pm • 0 0 • view
And yes, the stock market valuations of AI firms are scary high. www.cbc.ca/news/busines...
aug 14, 2025, 3:54 pm • 0 0 • view
They could easily build some special-case parsing to the chatbot so it can't be tripped up by easy word games, and the other features the older of the voice assistants had that made them actually useful, but that isn't bringing about the Machine Messiah so
aug 16, 2025, 9:30 pm • 0 0 • view
I'll note that even this faint efficiency is extremely dubious and what actual objective testing I've seen so far is that it makes AI-using employees *feel* more efficient when in fact they are greatly *less* efficient.
aug 17, 2025, 11:25 pm • 0 0 • view
Did they have AI replace the PR person who placed this story?
aug 13, 2025, 10:53 pm • 1 0 • view