2084: InstaImageGeneration, a really good Stack Exchange LLM, a way to generate designs from a text description, and a one stop shop

What a world we live in

Everyday I check my feed and theres so much cool stuff. First up, we have, by OpenAI of course, a new development called Consistency Models, which are essentially image generation models that allow one step instant generation. Now this is of course, a lot more real time than diffusion models which require a lot of steps to make all the pretty images, and it promises to allow a lot more interesting applications in the future - real time movie generation anyone? Generally they’re made by distilling down an existing diffusion model using the techniques they describe in their paper. They made their code open to use, well worth checking out. Currently it doesn’t seem to have any guided generation capabilities but I’m sure that’s just one paper down the line.

Next we have StackLLaMa, a variant of LLama, Meta’s open source large models, that has been trained on a ton of stack exhange data and as such, can answer all the questions you want in the style of a stack exchange post. It is pretty nifty and beyond cool.

Talking about RLHF and finetuning, there’s also DeepSpeed, which released their one stop shop for finetuning models. Some really cool stuff to check out there.

Finally, to save all designers everywhere, we have Galileo AI which automatically generates a design based on a text input. The video on the main page is super cool and well worth checking out. But it is in early access unfortunately.

I can’t wait for the future, it’ll be awesome. All these doomers about, who can’t see the potential of an endless stream of knowledge, available like one turns on a tap. The future will be wild and we are creating it day by day. You can’t turn aside from what’s to come, you can only dive in headfirst.