Bard vs. ChatGPT
27 Mar 2023 18:55:35 UTC
Topic 229308
(moderation:
I was only wait-listed for a few hours when I signed up recently for early access to Bard--which is Google's entry. I got ChatGPT access some weeks ago.
In just the last few days, I've been making a practice of cutting and pasting the identical text into each when I make a query. The questions are not designed test cases, but just things I genuinely want to know. Also, often, things that would not usually get me a single-query good answer in regular search, but require to walk down a path a bit.
My general reaction so far:
1. asking questions this way is often nice for me. I get a better appropriate level answer in fewer queries and keystrokes than I'm used to.
2. So far I am quite usually liking ChatGPT better than Bard for my purpose. The appropriate answer level and composition seems usually better, and on my small sample, the frequency of howling error seems less.
Unlike Mike's deep technical understanding questions for ChatGPT, my questions have been ordinary things springing from curiosity about my recent experience.
For example, I've just been re-watching the BBC mini-series "I Claudius" when I suddenly realized I was looking at Patrick Stewart (Picard from NextGen). I wanted to know what character he was playing, and whether that character had a historic foundation or was an invention of either Graves or the mini-Series. Both got me reasonable answers in a single try. ChatGPT's was a bit better.
For another example, we are taking a short trip next month with day stops in multiple North African ports. I inquired as to typical April weather in several of them and found them to be both cooler and more similar than I'd have guessed. Both gave good answers, ChatGPT's suited me better.
A lot of ink has been spilled on how this type of query will raise search computing costs, as it is claimed the providers are spending about seven times the compute per query as on regular search. But if my case is typical, people may get their answers with fewer trials. The bigger business concern is that neither one is generating direct hits on any websites, nor displaying any advertisements to me. So it is not obvious how the lost advertising revenue will be made up.
trust it a big problem with
)
trust it a big problem with both chatGPT and Bard. asking something you dont already know? unless you doublecheck that the answer is right, how do you know that the answer you were given is correct? and in that case, it didnt save you any time, since you spent time in verification anyway.
but chatGPT is much better than Bard. I've played around with both. it's faster and more comprehensive and coherent in my opinion. there were also several translation tasks I gave it (from Chinese to English) where Bard couldnt do it but chatGPT had no problem and translated right away.
I was also impressed with chatGPT's ability to understand pretty vague and subjective requests. I had a random thought yesterday and decided to see what would happen. "Who is the actress who looks like a younger version of <another actress>". and the response was good, it gave me several actresses who might fit the description, and the first one was actually the one I was thinking about.
_________________________________________________________________________
Did you guys watch the tv
)
Did you guys watch the tv show 60 Minutes a few weeks ago when they did a story on AI answering questions in real time on the taped show? One of the things the AI came up was that 3 INCHES of the ice in Antarctica is made up of "penguin urine" later on in the same show they fact checked the AI and it turns out penguins do NOT pee, so the AI lied or got it's fact from a non-reputable source and presented them as facts!! This leads me to believe Professors and Teachers are going to have to do ALOT of fact checking when people turn in papers anymore!!
Ian&Steve C. wrote:trust it
)
That is true and applies also to other 'knowledge' variations like Google, Twitter etc and yes, the beloved Wikipedia too. The, er, less sophisticated among us often have this notion that if it appears online it must be true. A low information tactic indeed, a sort of outsourcing of thinking, the ultimate in laziness IMHO. They have assumed, typically implicitly, the existence of some 'they' in cyberspace who vet information. So one can witness, as I have, apparently rational individuals who argue the merits of different points of view on some contentious topic by reference to cherry picked snippets from the internet : none of the participants realising they are all making the same error by such reliance on slanted selection of rubbish inputs. A good example of 'give someone a trowel and they can lay bricks'. It doesn't matter who you are, you can be wrong purely because your information sources are tainted, as frequently they are. Often this is justified in the name of freedom of speech, a genuinely risky strategy especially if physical truth is your goal. Consensus can be simple group delusion.
What to do ?
This comes back to scientific method, which appears to have been dumped for easier routes to enlightenment in recent decades. I urge you all to carefully read the following speech to Caltech students by Richard Phillips Feynman* ( self-cynic par excellence ) in 1974 : presented here. This is my 'cherry pick' for sure, mea culpa, but RPF has predicted much of the situation we are now in with what currently passes as science. I hope you will find it personally helpful, particularly with bodies of work that suffer from 'producer capture' - engineered bias masquerading as intellectual honesty.
Cheers, Mike.
* Who, while he was dying, incisively worked out the simple physical reason why the Space Shuttle went kaboom in 1986, and that later led to finding out the sociological reasons within NASA that made such a disaster inevitable.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal