Recent searches
Search options
Administered by:
I asked #Google Search a very basic, 100% fact-based question about the California DMV. The Google AI Overview confidently presented me with a 100% wrong answer. This is why I don't trust AI Overviews at all. I don't care how often they are correct, when they're wrong they're usually incredibly wrong and often in ways that could really screw people up who happened to believe those false AI-generated statements. And GOOGLE DOESN'T TAKE ANY RESPONSIBILITY FOR THIS.
@lauren I recently got this answer when asking a pretty simple question of how many active service members does the U.S. currently have
@lauren All that’s missing from this unsolvable word problem is, “How fast was Jane traveling?”
Last week tested ChatGPT-4o if it can provide me a formula for radio path loss calculation just to check what my students will get if (when) they'll use it.
Provided the right formula and I was, not bad. Then realized there was an error of three order of magnitude in both frequency and distance inputs. Would very easily have missed that if not had a wikipedia page open aside. (I'm CS and networking guy, RF as a side hustle when needed).
@jeaux @lauren
@puhuri Chat GPT should definitely come with a disclaimer… Unless dumbing us all down Dunning-Kruger style is the goal.
@benjistokman If it had just ended with the first sentence, I would’ve totally assumed it was accurate, but it just had to show off by adding something I didn’t even inquire about. I assume it’s 41% wrong—plus or minus, but I don’t know which.
@jeaux 41% decrease implies a 75% increase on the flipside. US population went up well under 50% in that timeframe.
@benjistokman I even thought maybe it was using the numbers 2023 and 1987 in its calculations, but that didn’t add up either. I finally just concluded that fractions are hard… even for AI.