While I expect this is fake, I don't know whats worse about it. the fact that copilot got that wrong, or the fact that they used it in the first place instead of one of the 3 built in functions that can summarize cells.
From what I can gather, it's not actually fake. Apparently you need to prompt the AI in a very specific way so that it produces the desired result instead of making the numbers up (which imo defeats the whole purpose of using the AI instead of learning the formulas)
My harkness test is that if you wouldn't trust it to run a nuclear power plant. Why on earth would you trust it making changes on your own computer?
"Help! There's a mode conflict at startup! ECCS pressure is plummeting and core temps are over 500°C"
ChatGPT: "Hmmm this is a complex issue that requires thought and consideration... Disable auto-SCRAM, insert rods, and pray to the god of your choosing"
I barely trust myself to run a nuclear power plant until the next scheduled maintenance (I wouldn't touch anything and wait for it to shutdown on its own), but I happily trust myself with root permissions and slapped together bash scripts (but not both at the same time).
14 Comments
Darkard@lemmy.world · 38 pts · 298d
Pika@sh.itjust.works · 11 pts · 298d
While I expect this is fake, I don't know whats worse about it. the fact that copilot got that wrong, or the fact that they used it in the first place instead of one of the 3 built in functions that can summarize cells.
black0ut@pawb.social · 20 pts · 298d
From what I can gather, it's not actually fake. Apparently you need to prompt the AI in a very specific way so that it produces the desired result instead of making the numbers up (which imo defeats the whole purpose of using the AI instead of learning the formulas)
https://office-watch.com/2025/copilot-excel-function-dangers/
marcela@lemmy.blahaj.zone · 17 pts · 298d
AI in Excel is simply a frightening idea.
irelephant@programming.dev · 8 pts · 298d
It's unfortunately real as well. https://programming.dev/post/39347864
Bongles@lemmy.zip · 6 pts · 298d
It's there and it's pretty bad
Avicenna@programming.dev · 14 pts · 298d
Lets put an AI button in the middle of the screen
BananaOnionJuice@lemmy.dbzer0.com · 11 pts · 298d
The parrot is rooting for the digital parrot.
trollercoaster@sh.itjust.works · 3 pts · 296d
But unlike stochastic parrots, real parrots can be taught to understand what they are saying to a certain extent.
ellypony@lemmy.world · 9 pts · 298d
My harkness test is that if you wouldn't trust it to run a nuclear power plant. Why on earth would you trust it making changes on your own computer?
"Help! There's a mode conflict at startup! ECCS pressure is plummeting and core temps are over 500°C"
ChatGPT: "Hmmm this is a complex issue that requires thought and consideration... Disable auto-SCRAM, insert rods, and pray to the god of your choosing"
anton@lemmy.blahaj.zone · 6 pts · 298d
I barely trust myself to run a nuclear power plant until the next scheduled maintenance (I wouldn't touch anything and wait for it to shutdown on its own), but I happily trust myself with root permissions and slapped together bash scripts (but not both at the same time).
pinball_wizard@lemmy.zip · 2 pts · 280d
If I followed that standard, I also wouldn't run Windows. And....I don't. So yeah. It checks out.
rook@lemmy.zip · 8 pts · 298d
Of course he gets the job, he is probably smarter than the whole team at MS.
Avicenna@programming.dev · 3 pts · 295d
I suspect parrots may be more sensible than that