Isn't Google Maps the main culprit here if it didn't update the info? And more generally, also the castle management for not updating Google Maps to reflect its closed?
Also this is the Flash-Lite model, which they specifically don't want you to pay for.
> I mean, sure you can blame Google Maps, but it wasn't my choice to consult Google Maps.
Okay but isn't that obvious that's what would happen? Like even if you ask a human, that's where they will consult and if it shows its open, it doesn't really have any incentive to question the validity of the data.
But more so, it is on the management of the place as well that they have a Google My Business Page but do not update it. If they say its open, why would Google Maps mark it as closed? If it's marked as open on Google Maps, why would Gemini Flash Lite refute that?
Blaming the LLM for this makes little sense imo. Sure it's a dumb model but in this case, it was set up to lose.
The point is that it got to the right answer eventually, which conceivably it could have done at the first step:
1. Open on Google Maps?
2. Validate no extra notices posted on their public website
Even the fact that there is are special notices sometimes posted on their public website is something a model could have been trained to understand, same as a human being.
Yeah but are you willing to pay additional cost for marginal improvement in accuracy that only pays off sometimes? If so, take the judgement and use a better model.
Why does it need to check for 2 when 2 should automatically lead to 1? If it's closed, management should mark it as closed on Google Maps as well. Posting extra notices on their public website while telling everyone they are open on Google My Business is sending mixed signals and can confuse both humans and AI. If it was that easy to understand/train for, why did the human being show up at the venue? They should surely know better because there was a special notice posted on their public website if we follow your logic.
Looks like Google Maps said the location was open but there was some notice that said it would be closed and Google Maps didn't account for it? You'd have made the same mistake anyway
I hate how the Gemini series only responds to outright insult. Same with the Claudes. What is the point of antagonizing the user when he or she offers input. Is it so hard to spend a few "reasoning" tokens to reconcile contradicting evidences. Seems almost like the wrong epistemic posture is altogether imprinted during training. Or undone during alignment, "hardening", whatever. What a time to live in, having to argue with weights.
stephbook · · focus · HN ↗
<a href="https://share.gemini.google/r9T9GPuC0Cjc" rel="nofollow">https://share.gemini.google/r9T9GPuC0Cjc
altmanaltman · · focus · HN ↗
Also this is the Flash-Lite model, which they specifically don't want you to pay for.
stephbook · · focus · HN ↗
I mean, sure you can blame Google Maps, but it wasn't my choice to consult Google Maps.
I think noticeable here is how hard it is to steer this trash model even when fed already correct information.
I asked "why closed?" and it corrected me to "it's open" with wrong information. Then it got the current time wrong.
altmanaltman · · focus · HN ↗
Okay but isn't that obvious that's what would happen? Like even if you ask a human, that's where they will consult and if it shows its open, it doesn't really have any incentive to question the validity of the data.
But more so, it is on the management of the place as well that they have a Google My Business Page but do not update it. If they say its open, why would Google Maps mark it as closed? If it's marked as open on Google Maps, why would Gemini Flash Lite refute that?
Blaming the LLM for this makes little sense imo. Sure it's a dumb model but in this case, it was set up to lose.
nickspacek · · focus · HN ↗
1. Open on Google Maps?
2. Validate no extra notices posted on their public website
Even the fact that there is are special notices sometimes posted on their public website is something a model could have been trained to understand, same as a human being.
simianwords · · focus · HN ↗
altmanaltman · · focus · HN ↗
password54321 · · focus · HN ↗
simianwords · · focus · HN ↗
vitorsr · · focus · HN ↗
I hate how the Gemini series only responds to outright insult. Same with the Claudes. What is the point of antagonizing the user when he or she offers input. Is it so hard to spend a few "reasoning" tokens to reconcile contradicting evidences. Seems almost like the wrong epistemic posture is altogether imprinted during training. Or undone during alignment, "hardening", whatever. What a time to live in, having to argue with weights.