DragGAN's Amazing Drag-to-Edit Photo AI Releases Source Code

0:00Today on the AI Breakdown, an update from the drag-to-edit AI image tool, DragGAN, a new LLLM from Inflection, and much, much more. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Welcome back to the AI Breakdown. This week, I am traveling in Europe. As you've heard, I've done a couple of interview shows, but for the next few days, I wanted to stay current with at least the headline news, so I'll be bringing you an abbreviated version of the podcast that is just the brief. Obviously, then each episode will be a bit shorter, but you'll still get all the most important headlines to keep you current with what's happening.

0:35One more note before we dive in. if you're liking the show and you listen on Apple, I would love for you to go drop a five-star rating. Believe it or not, a lot of people use the number of ratings as a benchmark for whether they're going to try out a show. So if you leave a five-star rating, it would make a huge difference. With that, let's get into today's AI Breakdown Brief. DragGan's mind-blowing AI image modification tool releases its source code. Inflection says that its LLM can compete at GPT 3.5 level. And a slew of products from YouTube to Google Sheets get an AI upgrade. Welcome back to the AI Breakdown Brief, all the AI headline news you need in five minutes or less.

1:12Today, we kick off with a product or really research that was released a few weeks ago and really captured people's attention. The research was called DragGAN, and it was basically a tool that allowed you to edit images by simply dragging them. Immediately, people could see that there were so many implications for this, and they really hit this idea that one of the superpowers that AI enables is to make anyone capable of so much greater, fine, granular tuning type of image modification than was possible before. Well, now DragGAN has released its source code, allowing people to go play with it.

1:46And there's even a test model that you can play with on Hugging Face. So I will include a link in the show notes and you can go start playing with that. Over in the world of LLM's Inflection, which is the company behind Pi, the personal AI chatbot, has announced its own foundational AI model. Inflection, which was founded by LinkedIn founder Reid Hoffman, as well as DeepMind founder Mustafa Suleiman, said that its Inflection 1 model is capable of roughly GPT 3.5, which is the base level of ChatGPT, size and capabilities. Now, when it comes to the comparison to ChatGPT-type platforms, Inflection 1 is competitive when it comes to things like middle and high school level exam tasks, as well as common sense-type benchmarks, but it really falls behind on coding, which is something that Inflection has said is just not a priority.

2:28Remember, the whole goal of Inflection's Pi is to create a type of companion AI that the founders believe will be an essential part of human experience in the years going forward. Next up, we have a massive set of new features from AI products, and we start with MidJourney. Now, I'm sure if you're watching this, you've probably already seen all sorts of information about MidJourney's 5.2 update with its big banner headline feature, which is Zoom. With ZoomOut, you can take a base image that you've created in Midjourney and do exactly what it says, zoom out. And that means you can put a single character in many different settings.

3:01Lots of people are really excited about this capacity to have a consistent character across lots of different types of environments. And there are also a ton of folks out there who are already figuring out how to use Midjourney Zoom to create a sort of cinematic film effect. Eleven Labs, who we recently talked about raising their Series A, has introduced a new feature that they're calling Voice Library. Basically, the way that this works is when you generate a new synthetic voice in 11 labs, you can then add it to this voice library. That gives other people who are using 11 labs the ability to tap into and use the voice that you've created, for which the user who created that voice will earn rewards, such as being able to create more characters for free.

3:38What makes this interesting to me is I think we're going to see a lot more of this generative AI community type product in which the things that people create with these tools become the basis for other people creating more with these tools in a really powerful self-reinforcing positive feedback loop. One update that I'm super excited about, YouTube is experimenting with dubbing in other languages via a team called Allowed. Allowed is part of Google's Area 120 incubator and basically creates a transcript of a YouTube video and then turns it into a voice dubbed in a different language. So maybe in one day not too long from now, you might be listening to this in not exactly my voice, but in Spanish or French or any other language.

4:14I think this could be extremely powerful for breaking down linguistic barriers when it comes to information and knowledge sharing. Google Sheets has announced a new AI upgrade called Help Me Organize, where you can enter a prompt and get a customizable table generated by AI. So for all of you Excel warriors out there, get excited. And LinkedIn, in conjunction with UC Berkeley, has developed a new method to detect AI-generated profile photos. In a recent test, the detection methodology correctly identified artificially generated photos 99.6 % of the time. It misidentified genuine pictures as fake only 1 % of the time.

4:47So in the fight against AI-driven scams, this could be an extremely powerful advance. Speaking of scams, zooming out to bigger society-level questions, Pew Research has just released a significant study in which experts predict the best and worst changes in digital life by 2035. I'm considering doing a full video about this, but some of the banner headlines include that 42 % of the experts said that they're equally excited and concerned about the changes in the, quote, humans plus tech evolution they expect to see. 37 % are more concerned than excited versus 18 % that are more excited than concerned.

5:18So overall, a bit more concerned than excitement. 2 % said they don't think there will be much real change by 2035. I'm interested to know what rock they are living under. Meanwhile, we've recently discussed how the UK is pushing forward with a number of AI safety initiatives, including their$100 million foundation task force, now being led by entrepreneur and investor Ian Hogarth. But you're also starting to see the government of the UK announce artificial intelligence efforts in specific sectors. For example, they've just authorized£21 million to explore how AI can help diagnose patients more quickly for cancer, strokes, and heart conditions.

5:49And finally, one more in Europe. In the weeks after the EU AI Act was passed, the CEO of NVIDIA says that the company is very likely to invest in Europe. Jensen Huang told reporters that expanding in Europe was a, quote, wonderful idea. Jensen said, I would say it's extremely likely, and the reason for that is because NVIDIA would like to be a global international company, and who could imagine a better place to invest in? given how essential computing power is to global competitiveness when it comes to AI. I'm sure the EU will be excited to hear that. All right, friends, that is it for today's AI Breakdown Brief.

6:19Check out the podcast and the newsletter. And until next time, peace.

6:39Bye-bye.