– Google New Imagen Text-To-Image Model
– The challenges and biases in large language models
– SNAP’s stock down 43% after missed earnings
– Advertising companies Meta, Alphabet, Twitter and more hit by economic fears
#imagen #imagendiffusionmodel #largelanguagemodels #llm #dalle #google #openai #alphabet #ganclip #drawbench #snap #snapchat #meta #twitter #adtech #texttoimage
All right hello hello, it’s Tyler Bryden here now want to acknowledge that I am filming this video the day after a horrible shooting down in Texas and one of those. Moments where. You both want to know and almost don’t want to know because of how heavy. This is and so. I’ve seen an outpouring of, you know messages and and videos and and.
You know people from all different perspectives, sharing their thoughts on this. I’ve I frankly don’t think I have too much of an informed. Opinion on it because. Of I guess even my both conscious and subconscious choice to not. Go that deep into the events of happening that have happened just because of how sad it is. But you know, thinking out there of you know anyone who you know. Obviously people who are the families who are affected by this. And then I think just the the world as a whole is always shocked by this by sometimes the.
Pure violence that people can be capable of, and. You know, I know everyone has these. These feelings, these pains, they we suffer through this and then we do our best to proceed with our daily daily lives and. Sometimes that’s easier than said than done. And I think you know, with events like this, the ripples of it last a long time and will last generations. For those you know especially affected.
Problematized by this and you know I talked yesterday about just like not operating from fear and. I think when something like that happens, it’s hard not to at least operate in fear for a little bit, and I feel for anyone who is sending their kids to school the next day after something like this and just praying for the best and praying that something like this doesn’t happen so. With that said, I’m going to move on to, I guess more of the normal things that I I talk about here and. Do my best at that so. One of the.
Sort of common themes through, you know, through these sort of stand ups. I think yesterday was 25. Today is 26 has been at least a tied thread with this concept of language and. Even since I’ve started writing or doing these videos, the amount of innovation that has happened is is truly it’s it’s insane, and it’s also this proliferation of people who are trying who have sort of hit this threshold of the amount of data they’ve been able to access. The technology that they have and all that has been unlocked.
So much powerful potential and then actual applications of this technology and so. Yesterday Google came after open AI and their Dolly system and you know, basically explicitly said that this system is better than than Dolly. Here’s some of the images that they that they released, so this is their image and it’s a text to image model and these were sort of. Obviously ones that they you know they did a lot of generation and then they shared the ones that were probably the most powerful and impactful. Yet there have been some people who have tested and played around with the system already and are seeing already sort of benchmarks about the. You know, the higher quality that this is outputting and they even did.
I have it somewhere they actually did. Some real, you know, some real benchmarks on this. Using human raters to then get them to, you know, share which images they prefer. So both in image text alignment and then image fidelity, and so with. With that, Google sort of systematically one one that I guess battle and it’s weird, right? You know you contrast this to the topic at the start of this this episode and the events yesterday won so dark and so.
You know, so just devastating for society. And then we look at a transparent sculpture of a dock made out of glass that was generated just by putting on a text. Just printing text and then and then creating that automatically. But you know, in a world that you know is again chaotic and and volatile. You know things like this, I actually think are a net positive and I think anytime that we can.
Describe something, describe the way that we feel, and then turn that into something I think is very, very powerful. And obviously there’s some very silly examples that have been created here, but I still believe I’m a believer in the the potential of creativity and art and expression that systems like this are going to create. And again, this is just this is just the beginning. This is only, you know, the Google’s really first release on this. Dolly, like the usage of these are are insane Wamba I being a company with at some points that the most downloaded app one boy allowing you to do some image generation like this as well too. And I think it’s just such a deep fascination that people have with this kind of technology and the ability that is possible here and one of the things I wanted to.
Quickly touch on. I thought it was really fascinating is that they were very clear about the limitations and the societal impact. And then also the you know their desire to not release this publicly as they still. You know worked through this and and you can see. I mean, there’s many reasons why, but really, the the main, I mean the main reason of this is that they are scraping large large data sets from the entire web, and that web is full of bias. It’s full of pornographic imagery, talk, toxic language, bias, racism, social stereotypes, it’s uncurated web data, so it inherits these biases and actually, you know, appreciate Google talking about.
Talking about this and allowing you to have the understanding that depending on the prompts that you put in that there are, you know there’s deep bias built into this status set and then. The images that you generate might have a reflection, so look, even this preliminary assessment has an overall bias about generating images of people with lighter skin tones. So and and then professions that align with Western gender standard stereotype. So I mean, I think it’s great for them to have released this and and talk about the bias and still do testing before this is released. Sort of publicly, and I’m sure at some point it will, and I think there will be private companies who actually want to release these things publicly, probably without the same thought.
Testing that has been done here because it’s there’s an incentive to do it. If they can be, you know, early in the market they’re going to have the ability to. Grow significantly and deliver on the desire for people to actually have access to these technologies and use them so one you know. Note for you if you do want to. Test this there is trying to think of a good example.
Yeah, I’m poodle dancing on a cloud with lightning. Strikes in the background. Filled with cats. I don’t know what this is going to do, but this is a on hugging face a space and is sort of the Dolly mini, but I believe this is the one that I’ve always used. It looks like they’ve updated the space and should generate a couple instances of this.
See, it’s taking a little bit to run here. There we go, a poodle dancing on a cloud with lightning strikes in the background, filled with cats. Definitely doesn’t have the same power and capacity of Googles and and and dally open AI system. But overall it’s still is achieving that goal. It’s nice that you see various outputs that come from it, and if you actually did want to play around with this technology, at least you have the ability to do some prompts and and get that live. So I think that’s sort of cool and just something I wanted to touch on.
And overall. You know you can see like even Tech Crunch talking about it. The the battle that is happening here with these large language models using the transformer approach and these companies wanting to one up each other on this. So there is definitely competition right now. It’s pushing AI researchers and engineers to the limit and I’m excited to sort of follow this battle. And you know I guess reap the benefits of it with I mean just some fantastic things. I love looking at these images. We’re moving towards video of this. There’s going to be whole.
Abilities in VR that are being built by this, it’s just going to be absolutely insane what actually, what, what takes place from this so. That’s one piece that I wanted to talk about today that is Google’s image in model large language model using the transformer approach to generate text to image. So definitely worthwhile checking out and then the last thing here as I’m coming up on time is this story from yesterday which if I can see one second here. The snap stock. OK here we go looking at a one day, but if we go one month it’s bouncing back a little bit today here. But yesterday 4 down 43% and really the message here is that the advertising industry isn’t going to be in a tough spot and that that this is sort of a cyclical cycle as inflation rises as interest rates go up as people look towards us in recession and everything that’s happening.
Advertisers need to rethink their spend. They need it to be directly attributable, measurable to the return on investment in the business. Those things around brand awareness and things might not be as valuable as they’re looking at cost cutting measures as they’re looking at the best ways to spend their money to generate revenue and then also conserve their budgeting so that they can last longer through a potentially. I know a long time of recessionary period where there could be challenges ahead that we don’t even necessarily foresee. So what was really, you know, sort of. Not shocking, but that went along with us was basically how other companies were then affected. So meta down Roku down Alphabet, Twitter down, Pinterest Down. So any because of what Snapchat sort of guidance was on this call. There was then just a.
Surrounding impact on any business that then touches that space. Primarily a lot of them driven by advertising models and related to text. So the ones Google Alphabet, Twitter, everything there and so. There is definitely another indicator of. A difficult time ahead, with SNAP being an example of that. Nivedita is Nivedita, Nuvita is has their earnings call today too, and so they’re there. You know we’re going to see what comes from that, but overall I think this next quarter.
Is going to be very. It’s going to be a huge signal for where we are at because last quarter you know this was starting to happen, but there was still still things OK and and really the environment has deteriorated very quickly over the last month. A month and a half, and so we’re going to see the downstream effects of that at the end of quarter two here, so. And that’s it for me today. I appreciate you tuning in and I know it’s, you know, it’s sort of a.
A dark time in in in the world. In many ways right now, so you know, we do our best to remain optimistic. We do our best to wake up. Drive forward with purpose, but the smile on her face and I hope you can find that ability here. And I hope that we can continue to grow and heal as as a world because.
You know all we can do is. Share these experiences together and hope that we learn and make better decisions and and and and grow together moving forward. So thank you as always for checking this out. Hope you have a good rest of the day and look forward to doing this again. Bye bye.