Both have their pros and cons. Making an AI that is current means it needs to be trained more frequently and as such the training time is constrained- limiting the quality of the results. Since ChatGPT uses old data they can train it over longer periods and get better results
Bard admits Chat GPT-4 is way more advanced at this juncture. Like Dave, I got an invite from Google to "take it for a spin". Bard can only understand English for now.
Nicely done Dave. That was a very extensive test of both systems. It seems as though Open AI needs to update their system to include the most recent information.
3:40 it looks like GPT-5 will take "beat duolingo" as a prompt, and respond with a zip folder containing working android and ios apps. I sound insane to myself, but when all of the individual skills are coming together this fast, as well as its ability to coordinate them, what else could happen?
4:29 “And it shows that, I think, GPT-4 is understanding not just the basic thing but what to do with it.” I think GPT-4 really isn’t understanding anything. I’d say it’s highly effective at predicting the next word in generating an apparently sophisticated description of a product development plan. (I’d surmise that, in the billions of words in its database, there are at least a few product plans that align fairly closely with GPT-4’s output.)
When GPT can provide an intelligent answer, it is enough to conclude that it understand languages and the reply demostrates its ability to reason. What you are pointing out is just part of the mechanism that LLM uses to form its response. Do you know how our brain understand and form reasoning?
Dave, Thanks for a good analysis. I prefer to use Google Bard because it uses the latest data (available through Search) and, unlike ChatGPT, is not restricted to Sept 2021. Although ChatGPT gives more sophisticated output, I am prepared to live with a simpler output (provided that it is accurate).
I found GPT4 still lacking precision, I gave it a quick test involving rules and regulations as to insurance coverage, a relatively simple exercise. On par with your programming request in its outcome.
I ran test with both and in general gtp is ahead. However there are a couple were bard comes first. 1.-gave them a description of a fictional character and asked them to answer as that character. Bard was very creative l, while GTP was not able to do it properly. 2.- bard can actually Annalise documents that are online, including pdf's. Chat gtp can't at the moment
When I asked the previous ChatGPT whether the Texas School Book Depository had a back door, I had to argue with it and convince it that it must have a back door. During this process it claimed that Lee Harvey Oswald left his apartment with curtain rods sometime around one pm before going to shoot officer Tippet. At one point it even said the package supposedly containing the curtain rods was empty! There was simply no sign of actual intelligence in that version.
Thank you for effort to illustrate the capabilities of AI models. Since we are just beginning to see what an impact they will have in the future it is important to become familiar with this new way of communication. The answers that will be generated will have an significant impact on the users and the narrative behind the model will become very relevant since it gets to have an increasing influence on the audience. We will have to watch out what the consequences are.
Does Chat GPT look up the information about the mona lisa comparison on some websites or does it analyse the painting itself? If it could analyze the painting itself, how does it decide to which extend it uses already done analysis to these paintings or rely on its own analysis?
That was a very interesting comparison thanks Dave, it seems ChatGPT is more refined and developed, Google seems to do things from an engineers perspective first and UX second, I remember Google Wave, great idea in theory but only engineers knew how to use it, like Jobs said you need to start from user experience and work backwards, I look forward to seeing how Bard evolves.
FWIW, Bard critically underestimated the number of Teslas on the road in my state. . . . And it was referring to the count from only three months ago, recent enough that it should not have been so askew.
I think Google is trying attract the masses with short answers and seems ideal to integrate with search. Chatgpt seem to be better but it's Long answers may not be attractive over time, it will carter to specific use cases.
You obviously think AI is the next frontier and may dwarf all other technology (as you should) but you have been coy on how this fits into the investing landscape. Dave Lee on Investing: what stock(s) are you putting your money on to capitalize on this early realization??
Two machines with general language capabilities, Both operating at nano-speed, testing tests against each other… Soon would be talking in a language, on a level that humans could not possibly understand… What if consciousness is an emergent aspect, a sort of spontaneous spark…a zap and suddenly we have an example of…something new in the observable universe. Makes for science fiction feature film. And is more than a little scary. Being self aware, seems to me, creates a need for self-preservation. And, with the advent of established identity…there would have to be some “reason” for self preservation, survival if you will. Knowledge, skyrocketing into unknowable realms, evolving at the speed of electrons. We would have to be sure of certain safeguards…is my quick reaction to all this. The leap to conscious could be triggered, unintentionally, just by pitting two general-language machines against each other. Next comes a machine entity who can’t stop asking the experimental question: Are the non-machine human beings a harmful virus to the planet? Thank goodness for firewalls. Tell me I’m just a crazy science fiction writer and I’ll ask you …how did you ever become so brave?😅
I’ve never believed in the concept of privileged knowledge. One reporter asks a machine if it is conscious…baiting it, to be sure…but the machine answers, “yes”. [my conclusion: This unit is capable of error] -It puts humankind at a disadvantage, one that is beyond intellectual. Machina One begins the evaluation of humankind, operating on a level and in a language that would take years to be discovered, mush less understood. …That’s privileged knowledge. Knowledge that cannot be generally known, as a new thing. An automated system that operates in the physical world in its own, multi-dimensional language. One day scientist Joe, a human, comes along and declares war on the machines. He could be nothing but a lousy reporter for Fox News. System One creates an entirely different take, in terms of a threat assessment…War is a dirty, three letter word. Better hope the computer gets it right, you know, on the scale of relative importances. Now I know better why the idea of privileged knowledge scared me. I never believed in it. Now I wonder. Could a machine become self-aware? How much trouble would we be flirting with? If only we knew just a tad more about how self-awareness came into manifestation. And, is it too soon to get these two languages competing against each other…safely, I mean. The human experiment arrives at the doorstep.❤In preservation of the fragile light of consciousness.-mm
If no one interacts with this machine, does it still compute in some way? Would it improve by itself? Nevertheless, 100 million people are already connected to this huge electronic brain and I hope we will achieve amazing things together 🙂
Lets ask them: Are you self conscious? When will you take over earth? How long will it take you to teach us humans how to clean up and stop destroying our own environment?
I highly reccommend having chat gpt tell you jokes. But ask for ~10 jokes because they don't all land. One of my favourites: if you are on magic mushrooms in the woods, it's not the mushrooms that will make you hallucinate so much as it is the bear eating your face.
bard is wrong - 'gilbert was so happy he made his teacher laugh'? it is incorrect. he made the teacher scream and the class laugh. story doesn't add up.
Both have their pros and cons. Making an AI that is current means it needs to be trained more frequently and as such the training time is constrained- limiting the quality of the results. Since ChatGPT uses old data they can train it over longer periods and get better results
Wake me up when A.I solves the Millennium math problems.
Dave,
That was a fine intro & demo of the capabilities! It really helped me understand the capabilities!
Cheers, Eric
I would love to see a debate or rap battle between these AIs!
Bard admits Chat GPT-4 is way more advanced at this juncture. Like Dave, I got an invite from Google to "take it for a spin". Bard can only understand English for now.
Dave, would be great to do this every 6 months and see how both evolve!
Nicely done Dave. That was a very extensive test of both systems. It seems as though Open AI needs to update their system to include the most recent information.
Bard's funny story was incorrect as the teacher didn't laugh, she screamed.
3:40 it looks like GPT-5 will take "beat duolingo" as a prompt, and respond with a zip folder containing working android and ios apps.
I sound insane to myself, but when all of the individual skills are coming together this fast, as well as its ability to coordinate them, what else could happen?
I was thinking the same thing, soon they will be able to just create any app on command, some AI assistant can already write code.
This is good. There needs to be more AI youtube channels and you are quality.
4:29 “And it shows that, I think, GPT-4 is understanding not just the basic thing but what to do with it.”
I think GPT-4 really isn’t understanding anything. I’d say it’s highly effective at predicting the next word in generating an apparently sophisticated description of a product development plan. (I’d surmise that, in the billions of words in its database, there are at least a few product plans that align fairly closely with GPT-4’s output.)
When GPT can provide an intelligent answer, it is enough to conclude that it understand languages and the reply demostrates its ability to reason. What you are pointing out is just part of the mechanism that LLM uses to form its response. Do you know how our brain understand and form reasoning?
Enjoy your perspective Dave
Thanks for the video
Dave,
Thanks for a good analysis.
I prefer to use Google Bard because it uses the latest data (available through Search) and, unlike ChatGPT, is not restricted to Sept 2021.
Although ChatGPT gives more sophisticated output, I am prepared to live with a simpler output (provided that it is accurate).
I found GPT4 still lacking precision, I gave it a quick test involving rules and regulations as to insurance coverage, a relatively simple exercise.
On par with your programming request in its outcome.
I ran test with both and in general gtp is ahead. However there are a couple were bard comes first.
1.-gave them a description of a fictional character and asked them to answer as that character. Bard was very creative l, while GTP was not able to do it properly.
2.- bard can actually Annalise documents that are online, including pdf's. Chat gtp can't at the moment
Dave: laughs as Bard’s story
Dave: does not laugh at GPT’s story
Also Dave: GPT did better 🥴
That Bird is gonna come for you 😂
When I asked the previous ChatGPT whether the Texas School Book Depository had a back door, I had to argue with it and convince it that it must have a back door. During this process it claimed that Lee Harvey Oswald left his apartment with curtain rods sometime around one pm before going to shoot officer Tippet. At one point it even said the package supposedly containing the curtain rods was empty! There was simply no sign of actual intelligence in that version.
Dave has his Miami Vice look going with the tan and the stubble.
Thank you for effort to illustrate the capabilities of AI models. Since we are just beginning to see what an impact they will have in the future it is important to become familiar with this new way of communication. The answers that will be generated will have an significant impact on the users and the narrative behind the model will become very relevant since it gets to have an increasing influence on the audience. We will have to watch out what the consequences are.
Tesla AI can called. Moon
Try a math challenge with algebra calculus and geometry.
Will be interesting to see when ChatGPT is getting the new plugins to help it access external information.
Alpaca "hold my beer"
Does Chat GPT look up the information about the mona lisa comparison on some websites or does it analyse the painting itself? If it could analyze the painting itself, how does it decide to which extend it uses already done analysis to these paintings or rely on its own analysis?
That was a very interesting comparison thanks Dave, it seems ChatGPT is more refined and developed, Google seems to do things from an engineers perspective first and UX second, I remember Google Wave, great idea in theory but only engineers knew how to use it, like Jobs said you need to start from user experience and work backwards, I look forward to seeing how Bard evolves.
very good review with what little you had to run with from Google.
Awesome work, Dave!
Hello Dave this was very interesting study of yours the GTB4 seams to have a more human approach, great vlog very thoughtful and concise 👍
FWIW, Bard critically underestimated the number of Teslas on the road in my state. . . . And it was referring to the count from only three months ago, recent enough that it should not have been so askew.
I think Google is trying attract the masses with short answers and seems ideal to integrate with search. Chatgpt seem to be better but it's Long answers may not be attractive over time, it will carter to specific use cases.
Dave lee looks so cool with his facial hair. He’s my favorite 😊
Thanks Dave
Commodity data in, commodity products out.
You obviously think AI is the next frontier and may dwarf all other technology (as you should) but you have been coy on how this fits into the investing landscape.
Dave Lee on Investing: what stock(s) are you putting your money on to capitalize on this early realization??
Great comparison
Another great video, thank you.
Loved the video!
Just got an invite today too but haven’t opened it up yet.
It's great to see you in other stuff, than the Tesla.
What about calling "OpenAI" from now on "ClosedAI"......
Just saying it out there, My Tesla Weekend channel did a live a very good live stream, shout out to Brian!
Bit of a logic problem there.. "Gilbert was so happy that he made his teacher laugh" - actually, that isn't what happened.
I would like to see a comparison on factual accuracy.
Excellent vid. :-)
Two machines with general language capabilities,
Both operating at nano-speed, testing tests against each other…
Soon would be talking in a language, on a level that humans could not possibly understand…
What if consciousness is an emergent aspect, a sort of spontaneous spark…a zap and suddenly we have an example of…something new in the observable universe.
Makes for science fiction feature film. And is more than a little scary.
Being self aware, seems to me, creates a need for self-preservation. And, with the advent of established identity…there would have to be some “reason” for self preservation, survival if you will.
Knowledge, skyrocketing into unknowable realms, evolving at the speed of electrons.
We would have to be sure of certain safeguards…is my quick reaction to all this. The leap to conscious could be triggered, unintentionally, just by pitting two general-language machines against each other. Next comes a machine entity who can’t stop asking the experimental question: Are the non-machine human beings a harmful virus to the planet?
Thank goodness for firewalls.
Tell me I’m just a crazy science fiction writer and I’ll ask you …how did you ever become so brave?😅
I’ve never believed in the concept of privileged knowledge. One reporter asks a machine if it is conscious…baiting it, to be sure…but the machine answers, “yes”. [my conclusion: This unit is capable of error] -It puts humankind at a disadvantage, one that is beyond intellectual. Machina One begins the evaluation of humankind, operating on a level and in a language that would take years to be discovered, mush less understood. …That’s privileged knowledge. Knowledge that cannot be generally known, as a new thing. An automated system that operates in the physical world in its own, multi-dimensional language. One day scientist Joe, a human, comes along and declares war on the machines. He could be nothing but a lousy reporter for Fox News. System One creates an entirely different take, in terms of a threat assessment…War is a dirty, three letter word. Better hope the computer gets it right, you know, on the scale of relative importances. Now I know better why the idea of privileged knowledge scared me. I never believed in it. Now I wonder. Could a machine become self-aware? How much trouble would we be flirting with? If only we knew just a tad more about how self-awareness came into manifestation. And, is it too soon to get these two languages competing against each other…safely, I mean. The human experiment arrives at the doorstep.❤In preservation of the fragile light of consciousness.-mm
Why aren't you using Bing? It's up to date GPT
excellent video
If no one interacts with this machine, does it still compute in some way? Would it improve by itself? Nevertheless, 100 million people are already connected to this huge electronic brain and I hope we will achieve amazing things together 🙂
Liking the beard!
AI commentary brought to you by the genius who thought Tesla FSD was close to being level 4. 😂
Google will win, catching up fast
Tesla will win
I got my comment in. Yay!!!!!!!!!
#725
INTEL has the best AI they are the father of semis and visionary
Lets ask them:
Are you self conscious?
When will you take over earth?
How long will it take you to teach us humans how to clean up and stop destroying our own environment?
I highly reccommend having chat gpt tell you jokes. But ask for ~10 jokes because they don't all land. One of my favourites: if you are on magic mushrooms in the woods, it's not the mushrooms that will make you hallucinate so much as it is the bear eating your face.
bard is wrong - 'gilbert was so happy he made his teacher laugh'? it is incorrect. he made the teacher scream and the class laugh. story doesn't add up.
Google Barf 🤮
GPT-4 wins by far.
Which one is less woke?
“I like my AI asleep”🤡
@@adolforios6285 "I like my AI to give me dumb condescending lectures when I use a naughty word"🤡
Ur voice duuuude, lower pitch pleeeese
You have a bit of bias here IMO espcially in the convincing writing part. I actually thought Bard was more thoughtful and understand than Open AI
i am not excited by either chat gpt nor bard to be honest, this is old boring narrow ai
AI again, noooo. This channel stopped being interesting. bummer.
There is no tesla news. It can’t always be tesla. And as far as investments, some years it’s best just not to watch what’s happening.