ChatGPT got it right on the second attempt. I'm convinced Bard just makes everything up.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
AI probably some way away from ruling the world
- Thread starter Belly Buster
- Start date
ChatGPT got it right on the second attempt. I'm convinced Bard just makes everything up.
kelvin 660
White SR+ with LFP battery
Still wrong!
There is approximately 44 kWh in a UK gallon of petrol. Therefore 44 ÷ .3 Wh = 146.6 mpg
There is approximately 44 kWh in a UK gallon of petrol. Therefore 44 ÷ .3 Wh = 146.6 mpg
Cardo
Active Member
Why are you applying the low efficiency of an ICE vehicle to an EV when calculating the MPGe?A car doing 0.3wh/mile would be impressive!
Wikipedia states petrol 33.4? kWh. Us to UK gals X 1.2 = approx 40 kWh. But petrol cars only about 25% efficient so useable power 10 kWh so Tesla equivalent 33.3mpg
According to my dubious math
I dunno, that seemed pretty impressive to me. Might not have given the right answer but given the fairly esoteric question, and then the follow-up which is completely devoid of punctuation and difficult for even a native English speaker to read, it could have been worse.
Also worth bearing in mind that none of these AI tools are designed, as yet, to cope with mathematical operations. It’s not part of the training and getting a spontaneous answer that’s correct is emergent behaviour.
Also worth bearing in mind that none of these AI tools are designed, as yet, to cope with mathematical operations. It’s not part of the training and getting a spontaneous answer that’s correct is emergent behaviour.
MrBadger
Badger out
GIGO. Let's just hope that they can remove the garbage data sets from all the shite they are training various things with. At the moment, it sounds like they are just hovering up any old data, irrespective of the quality of the data. I would hate my car to have learned to drive in a way that I see some cars being driven.
They can handle this. For instance if you train it on a data set from kids it would learn that Santa and the Tooth Fairy is real. But with all the data from a adults as well it should learn when to say when Santa is real or not.
It also shouldn't think the earth is flat even if some humans do.
Not all perfect yet of course but it's impressive already.
But it's not really _that_ esoteric a question. It's unit conversion.
<UK gal per US gal> x <kWh per gal US> / <Wh per mile> * <W per kW>
~= 1.2 x 33.7 / 300 * 1000 ~= 134.8
You can get the rough 1.2 from <UK floz per US floz> x <UK floz/pint> x <US floz/pint>. US floz is about 4% bigger, UK pint has 25% more floz. It's interestingly close to 1.2.
The problem is they never will..it'd cost too much. Hiring a load of people to go through millions of data sets to provide clean accurate data (and proving that those people wouldn't introduce biases themselves) is an insanely expensive endevour.
Crawling the web and picking up any old crap? That's cheap.
It's getting worse too - because there are a load of GPT generated pages out there (especially marketing and fake review pages) they're being trained on their own garbage output..
Similar threads
