Most people who tinker with machine learning technology hope that their work will be seen by others in their community. Boris Dayma, the creator of the artificial intelligence called "DALL-E Mini," was shocked when it went viral.
The French-born machine learning expert tells us that he's mostly happy to see, after a chaotic few weeks and one name change.
The conversation has been edited to make it clearer.
Is DALL-E Mini related to Openai's Dall-E?
The DALL-E was the first model that was able to make images from text that were pretty good, but it was not public. You couldn't use it. The open source community tried to duplicate it.
When you build a model that is inspired by another model, you usually add something to it. It's not uncommon. There is a model called BERT that is built on top of an open-sourced model called BART. It is almost an homage to the first model.
DALL-E Mini is a small version of DALL-E, and I wanted it to be used by everyone.
I've just done the most famous project that uses the DALL-E name, and there are hundreds of others. There were many others before me.
Did you change the name because Openai was angry with you?
Nothing like that. It was time to change the name because my model had been out for a while and it was time to get back to work. I don't think asking to change the name of the car is the best way to go about it.
F: What's the reason your model went viral?
The first version was released at the end of July. It wasn't that great, but I was happy. You couldn't do very complex things if you input on the beach under the sun.
I've been working on making it better. I think it reached a level of quality where it became popular. From that time, it had a digital audience, but also in the artificial intelligence community. I've been talking about how I built it. There were people who liked to play with it, and a lot of people who knew about it. I think it's reached a point where I was able to create things that I didn't know was possible.
I think people just said "oh, it's good" and it became popular.
Even though nobody has dreamed up a monster like the one the network created, are you aware of it? Are you aware of what happened there?
The one who drew the prompt was behind the scenes. I was not aware of it at all.
The account, which is called Weirddalle, is quite impressive. Some people come up with crazy ideas. Some of them have a trail cam of teddy bears. It seems to look good. It's great.
What do you think about the content?
Some of the early models weren't produced by me. You might ask it something very innocent and it will produce some pornography. I attempted to stop that from happening.
It seems to be working well so far.
It came out surprisingly well, like of Richard Nixon snorting lines of coke in the Oval Office, thanks to the use of the artificial intelligence. Is it possible that it takes those leaps to create such strange and spot-on images?
It's kind of impressive, even for the basics, like putting the Eiffel Tower on the moon. I am still amazed at that. The thing works well.
What is your favorite prompt?
I just follow that account. There is a lot of creativity there. The trail cam images are really nice.
What was it like for you when the model went public?
It grew quickly and became large. I was trying to improve models. There was a lot of people who wanted to use it and I was busy doing that. I would have to speak to the reporters. I think it changed my routine a little bit.
I had to put my new model on hold while I tried to fix the server because it was too busy when it became famous.
Overall, I'm happy that the reception is positive and people have a good reception.
What are you doing right now?
I want to make these models better so they reach their limits. I'm not sure what I'll do next, but I like media generation. There's some work that can be done with videos, creating videos, or maybe creating sound or music, that I think can be useful.
There's more on the dark side of Craiyon.
All Rights Reserved © 2024
