Dear Fellow Scholars, this is Two Minute Papers
with Dr. Károly Zsolnai-Fehér. Dr. Papers. Today we are living the age of AI chat
assistants. That is not by accident, these assistants are incredible, and with
browsing support, they can find deals for us if we wish to buy a pair of headphones,
play text-based games with us that we make up on the fly, they are excellent at
coding and can even pass the bar exam. Now I have heard some folks saying
that Google is so far behind on this, and today you will see that this is not
the case at all.
Google is coming with guns blazing. Here is an incredible demo
of Bard, their ChatGPT-like large language model assistant. So, what can it do? Is
it as good as ChatGPT? Let’s see together. It can, for instance, code up chess moves in
python. Although it’s python, so the task is now as easy as it gets, it is nice to see that
it knows all the relevant libraries and concepts to make it happen. But there is something super
interesting happening here. Look! It will tell you what the output of this piece of code will
be. Something that ChatGPT is not that willing to do.
And it gets better – it also cites its
sources, even when it comes to writing code. Then, when done, we will be able to move
this Colab, an online notebook for running your coding experiments. And it also has the
favorite of the Fellow Scholars, dark mode. So these assistants are mostly limited to text
information. For instance, we can ask ChatGPT to recommend us text to image prompts, and the
results will often be fantastic. But it can not generate an image, at least, not directly. To show
how powerful this concept is, we will now take these ridiculously detailed scholarly prompts that
really embody the spirit of Two Minute Papers, and to go a separate text to image AI and ask
it to generate these scholarly images. First, Illustration of a computer program
rapidly sifting through scientific papers, extracting key findings and condensing them into
a two-minute format. Loving it. Very imaginative. A digital representation of a sentient AI,
depicted as a scholar, surrounded by floating abstracts of research papers with a timer in the
background counting down two minutes.
Artwork of a futuristic AI professor delivering a rapid-fire
lecture about a complex computational topic, with visual aids and diagrams materializing
from a holographic interface. And he is doing it with style. So good! And these prompts
are just getting more and more ridiculous. Get this, now, a visual metaphor of a dense
scientific paper being wrung out like a wet towel, with the drips forming simplified, easily
digestible information nuggets. Wow! And it still delivers! And if we so desire, we can ask
for 10,000 variants of these concepts and it will do exactly that, and very, very quickly. So just
imagine how powerful this concept of having large language models be able to create images would
be. But the assistant itself can not output or even show images directly. Until now! If we ask
Bard to recommend sights to see in New Orleans, oh my. What do we have here? Look! Images are
part of the recommendations too. Note that these images are found, and not generated, but I am
fairly confident that this is going to happen, and quickly.
But wait, it gets better,
what is that? Is it possible that? Oh yes, that is an image input, which we can now provide
to Bard, and it will do whatever we wish, in this case, it will write a funny
caption for it. Or as many as we wish. And as a good Fellow Scholar, if we
wish to get back and learn some more, and perhaps go to college, and
we are interested in video games, it can recommend colleges that offer this kind
of programme, and that is not so surprising. The surprise comes now. And now it fires up
Google Maps and shows them to us on the map. And this is just a small example that showcases
the true potential of these assistants.
And that is, integration with other programs. And if
we wish to cram all this information into a beautiful little table, we can do that too.
But wait, ChatGPT can do this as well, so why is this interesting? Well, hold on to your papers
Fellow Scholars and look! We can add an additional column to it on the fly describing whether these
colleges are private or public. And here you see one more instance of good software integration,
where we can add it to google sheets and share it with our friends with just a click. However,
the integration is not limited to Google’s apps, you will be able to infuse this assistant with
the power of Instacart, Spotify, and my favorite, Khan Academy in the next few months.
Our next
video is going to be on exactly that the work is beyond amazing, so make sure to subscribe
and hit the bell icon if you’re interested. But the list does not stop there. Remember that I
said that it would be absolutely amazing to have the AI write prompts and generate the images in
one go? Look. Oh my! Adobe Firefly integration! Yes! That is exactly what is going to happen
soon. What a time to be alive! And it gets better, Bard is free to use as of now, and is available
in 180 countries as of the making of this video. And finally, there is now competition in this
space, which is really good for us, the users. And don’t forget, all this is possible because of
the existence of these amazing research papers. In ChatGPT’s case, GPT, and in Bard’s case, a work
called Palm. So once again, everyone is talking about the Papers! Some might not know about it,
but they do. And just imagine what we will be able to do just two more papers down the line.
My
goodness. So, for now, let the experiments begin! Thanks for watching and for your generous
support, and I'll see you next time!.
