Skip to content
Gemini: The Most Profound Technological Shift in Our Lifetimes episode artwork

Dec 8, 2023 20 min

Gemini: The Most Profound Technological Shift in Our Lifetimes

from Search News You Can Use - SEO Podcast with Marie Haynes Hosted by Marie Haynes

In this episode, we dive deep into Google's revolutionary new language model, Gemini, destined to be the most impactful technological shift of our time. Host Dr. Marie Haynes unravels its capabilities, which extend beyond text to encompass images, video, and beyond. Discover how Gemini's potential to transform information access, problem-solving, and decision-making could reshape our lives. While we wait for its full potential to unfold in 2024, learn how Bard, soon integrated into Google Assistant, offers a taste of the future. This episode delves into the criticisms of AI, encouraging listeners to actively engage with Bard and explore its possibilities. We also explore the exciting world of GPTs (apps built on chat GPT), which may possibly be seen in a similar form in Google AI Studio.

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

Gemini: The Most Profound Technological Shift in Our Lifetimes

0:00 20:29
of matches

Transcript Auto-generated, may contain errors

Welcome to another episode of Search News You Can Use with me, Dr. Marie Haines. While it finally happened, Gemini has launched. Google has announced that Gemini is live.

It's a part of BARROT now, although BARROT doesn't have the full capabilities of Gemini, which is confusing a lot of people. I thought in this episode that I would just quickly tell you what Gemini is, kind of trying to answer some of the common questions that people have. Gemini is a language model. Just like with chat GPT, GPT is the language model.

It's the model that's been trained on text to understand the relationship between the words in that text. But Gemini understands much more than just text. Now, you're going to see people saying that BARROT is bad. I've already had people send me examples of hallucinations.

There's still some hallucinations. We're going to talk about that in a minute. Still, Google says that Gemini is the most profound technological shift in our lifetimes. That's a really big deal.

What makes Gemini radically better than what we currently know is what they call multi-modality. It's a little bit of a hard concept to grasp how we go from a language model that can write us a poem to something that can radically change the world, like I keep saying is going to happen. I'm going to do my best in this episode to try to help you to see what I'm seeing in terms of how this technology is going to change the world. Let's start by first talking about what a language model is.

It's just exactly what it says. It's a model of language. Gemini is trained on understanding not just language, but also images and video, possibly sound as well. Google published this technical report that says that Gemini is in its current state, Gemini Pro that doesn't have the multi-modality is just as good as chat GPT 3.5, the free version, perhaps a little bit better.

If you're using free version chat GPT, then Bard is better. You're still going to get some hallucinations. It's still not perfect, but it's better. Now, some, and then Google said that it's actually better than GPT 4 with the multi-modality.

Some people in academic AI circles are debating Google's claim there. We don't know. We haven't seen the multi-modality yet. There are three versions of Gemini.

Nano is for your phones. Pro is for text, and that's what we're seeing in Bard right now. And then Ultra also incorporates the multi-modality. Ultra is expected to be released in early 2024.

Right now, we just see a little bit of the power of Gemini. Ultra apparently outperforms GPT 4 on many benchmarks. People are saying that it's on par with GPT 4. That's the paid version of chat GPT or better.

I expect that as we see Gemini Ultra be launched, that it will get dramatically better faster. It uses AlphaGo's learning capabilities. AlphaGo was created by DeepMind from Google as well. And one of its superpowers is that it is really good at learning.

Bard will continue to improve regularly. But we still haven't got to the point on how that changes the world. How does a language model that many find not that helpful at all today, although I find it increasingly more and more helpful, actually change civilization? I think it all comes down to reinforcement learning.

And I'll tell you I have a lot to learn here. This is going to be one of my focuses over the next few months, is to understand more about how reinforcement learning works. We've talked about it in the context of Google's quality raiders. The quality raiders use Google uses their results.

They label results as helpful or not helpful. And that helps to train machine learning systems so that they're more likely to predict pages for Google's case to predict pages that are likely to be helpful. It's a model that continually gets incrementally more and better with each time that it goes through a training cycle. So reinforcement learning with human feedback, the human raiders, has helped search improve over the last decade, or however long Google's been using it.

But it takes time to do that. But what if you could use the language model, so Bard or Chat GPT, to actually do the quality rating? And I know that sounds like a scary prospect. There are a number of papers that talk about it.

I asked Bard for some papers. And it took me to a paper that's talking about a type of reinforcement learning that's called self-play, where the algorithm learns by playing against itself. And it doesn't require any direct supervision, doesn't require human input. I think that what Chat GPT and Bard do is a combination of the two.

But they're using the language models themselves to figure out, oh, that was helpful. That was a good result, or that was a bad result. It wouldn't surprise me if for every single conversation we have, it gets analyzed by a language model itself to figure out, was this a good answer? Was it satisfying to the user?

Was it a frustrating conversation? If so, then perhaps they use that in the reinforcement learning algorithms as an example of a conversation that wasn't good. And the whole time, these algorithms are learning to get better and better, to get more and more accurate, more and more helpful. This right now is the worst that Bard will ever be.

One day, we're going to tell our grandchildren about the days when Bard was so bad and got things wrong. It's already remarkably better than it was when it first came out. And I just keep saying this because people keep sending me these examples of, oh, no, Bard hallucinated. You're going to expect that.

I treat Bard and Chat GPT as if they're my assistants in a Slack channel. They're not going to get absolutely everything right, but they're going to be very, very helpful. And if you can learn how to dig into that knowledge, the depth of knowledge that Bard and Chat GPT have, how to get into deep into their vector spaces, then there's so much power that can be used in there. But we can't expect it to be perfect.

And that's something that I think holds a lot of SEOs back from trying to use these tools to do more in their lives. So multi-modality changes things. Now, instead of learning about how to improve text responses, when Bard uses or Gemini uses reinforcement learning, it's learning bit by bit how to improve its understanding of everything, of how things interact in the world. And when it's gotten its understanding wrong, I want to learn more about fine tuning.

Fine tuning is where you give a tool, a system like this. You give it an example and you say, this question, here's a question, and here's how I would like you to answer it. Here is another question. Here is how I would like you to answer it.

And what I believe you can also do is say, here's a question that you answered improperly, and here's a better answer to that. Now, that takes ages for humans to write out. But again, language models are becoming able to do this. And so that makes learning go incrementally faster.

Every time there's a training run, every, I actually think now more, I think about it, I think when we have core updates for Google, that that's just another training of Google's model that determines which content is helpful on the internet. There's a video Google released of a demo of interacting with Gemini, and it's getting a little bit of flack because it actually was edited a little bit. I still think it's an incredible thing, and I don't want to get into that controversy. I want to focus on this one part almost near the end of this video that really, really got to me.

I'm going to play the audio, but what's happening here is they show a picture of a cat that is sitting on the ledge looking up and about to take a jump. What do you think might happen next? The cat is going to jump to the wall and stick the landing. It's going to be a perfect 10.

Oh, and it's a miss. It looks like it was just a little too far away, but don't worry. The cat is a natural athlete and it will be back up there in no time. How did Gemini know that the cat didn't make it?

I asked Bird about this, and Bird told me about a Google paper on something called Open Scene 3D. I don't know if that's behind Gemini, but it was really interesting to read. This was a paper that talked about identifying objects and where they were in rooms and where they were expected to be based on a frame-by-frame looking at a video. And so Gemini not only can understand text in the transcript of a video, but it probably can understand what objects are in the video and where they're expected to be and where they are moving in a video and other attributes about the video itself.

I think it can determine emotion, perhaps. So Gemini is getting a more and more robust understanding of the world. What's next? It was a wired interview with Demisisovis, co-founder of DeepMind, and he said that Google is now looking into how Gemini might be combined with robotics to physically interact with the world and to touch, they have touch and tactile feedback.

Robotics that understand and perform actions. Now we're getting a little closer to something that changes the world. Although, I mean, it's exciting, perhaps I have a robot that does my dishes and my laundry and solves those problems that I have in my life. That certainly helps me, but it's hard to imagine the way that new machines will be created, the way that we can interact with our world with this technology.

It will certainly change the way that we live. Here's what I think is going to change the world. I've been talking for the last few weeks about GPTs. If you're not familiar with GPTs, they're a part of chat GPT.

We've only been around for a couple of weeks now. And what they are kind of like apps, you can create an app just by talking to chat GPT. I can say something like, I can give it a file, like maybe the quality reader's guidelines and Google's guidelines on helpful content. And then I can use natural language to say things like, I want this GPT to help me improve my content based on the information that is in these documents.

And I can give it even more instructions, like when the user asks this, respond in this way. GPTs are a way to program with natural language. Programming right now is limited to people who are programmers. And there are some very, very intelligent programmers in the world, but there are a lot of very intelligent people who don't know programming and are limited in what they can do.

So GPTs changes that. Well, Google's announcement of Gemini was quite long. When you get past all of the information on the benchmarking and all of the technical stuff, they have this little thing about something that they call Google AI Studio. They say that it's a free web based developer tool to prototype and launch apps quickly with an API key.

And then they say, when it's time for a fully managed AI platform, you can move it over to Vertex AI, which is Google's cloud platform. You know that I've been really excited about what Microsoft is doing in allowing businesses to essentially create their own GPTs, but with their own private servers and private data. This is one of the problems with GPTs is that you're basically giving your information to train open AI systems. And having your own private privacy is really a big deal before this can be something that companies actually use and make things with and make money with.

But it's not hard to see that once you have your own private servers, there's a lot that can be done. Any company that has a knowledge base that has processes, those can all be written as prompts. It's going to change how we interact with information in the world. DeepMind also has something called alpha code.

And this is something that helps people to produce code just again based on natural language. When I say natural language, I just mean talking, just doing what I'm doing right now. You can read all sorts of guides on how to prompt chat GPT and how to prompt bard. The best thing is to just talk to it.

When you get a result that isn't good, ask the tool. I didn't like this. This isn't good. I wanted to see something that had a chart with a line on it or whatever and talk to it.

Don't try to look up a solution. Just talk to it to get the solution. So alpha code is something that is incredible. I spent some time reading the paper for alpha code.

It uses several Gemini based models. Together they figure out the code that's necessary to solve a problem. And then it critiques its own code and removes the parts that's redundant. And comes up with the best solution.

Now it's not perfect, but it's going to continue to improve as well. We're just in the very early days of this. So you're going to see a lot of criticism of bard. You're going to see a lot of criticism of AI.

Think about why you see criticism of AI. It's because these things are a threat to media. The media that you're reading this criticism on doesn't like AI because it's going to impact their bottom line. The SEOs, many SEOs who I love and I respect your opinion, but it saddens me to see you criticizing bard.

I know that bard's got some issues, but I feel like criticizing these language models is like if you found a talking dog, but some of the things it said weren't exactly right. And that's what you focused on. The models bard, the tools bard and chat GPT, they have such a deep understanding of the world. What we need to do is figure out how to talk to it to get that understanding, to make it useful to us.

So for now it's kind of fun. You know, bard didn't, bard saw that the cat didn't make the jump. That's not going to change the world. But GPTs and apps, Google AI studio, I believe it's going to let businesses create some really, really creative things.

And because that's driven by money, that's going to allow businesses to make money, commercialism drives innovation. We're going to see some big changes and I think that that's going to probably start next year. It's going to take a while for us to learn the skill. One of the things I did with GPT is make a Fortnite coach and it takes some advice from some of Fortnite pros and then I can talk to it about things that I'm struggling to do.

And that's great. But I asked bard, can you see my Fortnite video? Now now it can't. But it told me that once it has multi-modality, bard can see the frames in my video.

It can analyze my gameplay. It can compare my gameplay against competitors. It can find my weaknesses. It can tell me where to improve.

It can help me work on goals. It could create a coaching program for me. If bard could do that for my video game, what happens to athletes? Analyze every frame.

Find your weaknesses. Compare against your competition. Determine what exercises are going to help, what supplements, whatever's going to help you improve. I don't know what happens to humanity when we optimize every single aspect of being a human.

Gemini is going to help us solve problems. Google talked about how researchers can use Gemini to look at papers and not just collate find where papers that talk about a particular topic, but actually where the images do, where the charts do, where the relationships are. Scientists will be able to develop new compounds really, really quickly, perfect new techniques, whatever problem it is that there are people who have dedicated their lives to solving the world's hardest problems. All of a sudden those are radically easier now.

So that power is in bard. Again, we're not going to see the full use of video and images and the full power of Gemini until next year. I have a challenge for you. I want you to use bard as much as you can.

I have nothing to gain by saying this. I just see it as the way of the future. Your goal should be to get to see what you can get it to do rather than where it fails. Bard is soon a part of Google Assistant.

It'll soon be silly to think that when we wanted to find something, when we wanted to know something in the world that we'd type in keywords, we'd navigate to a web page, we'd scroll around until we found the answer, try to get past the ads, read the answer, but it's not quite the full answer. Go back to search, do another search, go on another page, scroll around until we find that. Maybe click on a few things in order to get the answer to our question. Bard will eventually just answer all of our questions.

I know that's scary. I have people reaching out to me now saying, I love your podcast, I love your writing, but I'm scared. Sam Altman, the co-founder of OpenAI said that every 100 to 150 years, half of the world's jobs are lost. I think we're going to see some big, big changes.

If I'm right, Gemini Ultra will make a huge shift in humanity. Google said it, the most profound shift of technology that we're going to see in our lifetime. So yeah, there's some things to be scared of. We don't like change and we don't know what the future's going to be.

But I would say that if you're listening to this podcast, you're already so ahead of the curve. I'd encourage you to become a Bard expert, not a critic. The more you understand how to work with it, how to get helpful information from it, the more you're going to have advantages.

Frequently asked questions

How long is this episode of Search News You Can Use - SEO Podcast with Marie Haynes?

This episode is 20 minutes long.

When was this Search News You Can Use - SEO Podcast with Marie Haynes episode published?

This episode was published on December 8, 2023.

Is there a transcript available for this episode?

Yes, a full transcript is available for this episode. You can read the complete transcript on the episode page.

Can I download this Search News You Can Use - SEO Podcast with Marie Haynes episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.