Unbelievable (watch the examples towards the end for maximum amazement). https://t.co/eePbOgnOYv

I maintain that @wavymulder’s Analog Diffusion (a fine-tuned version of #stablediffusion) is one of the best models ever released so far. https://t.co/EMoQGl3KS3

A wild dream I have: a LLM so sophisticated (imagine the great-grandchild of #copilot or #chatgpt) that I can say:
“Take this open source app and add feature X. It must look like this and return the results like that.”

The #AI would modify the source code to blend my feature in.

Also, shout out to @chidiwilliams__ for creating an open source implementation of #whisper, Buzz, that uses both @OpenAI original models and @ggerganov optimized models.

https://t.co/LbtBUwcdbT

I am convinced that if the world gets an open source version of #ChatGTP that can be deployed locally, that runs on consumer hardware, and that is easily fine-tuned just like #stablediffusion is, there will be an explosion of apps like we have never seen it.

Not just by improving the inference performance of #whisper on Apple silicon, but also in showing its potential with very impactful demos. For example, his “GPT-2 meets Whisper in WebAssembly”:
https://t.co/L9mRuuygut

Put aside the hypothesis that “Google is done”.
Put aside the complete unreliability of #ChatGPT.

The implication of this is that people won’t visit website anymore. And if they won’t, website owners won’t earn from ads anymore, won’t get a chance to build communities, etc https://t.co/1a4Xcpq9eF

Very impressive. https://t.co/iymSgNOjdv

@camenduru There are plenty of other ways to achieve that without impacting discoverability. Text like “The AI Model Archive – All creators are welcome.” or something like that, strategically positioned on the page, will help you.

This feature made me think about a(n absurd?) idea: imagine if @YouTube would expand the number of livestreamers to XY and it would allow private streams for them. Ad-supported office web meetings. The ultimate nightmare.

https://t.co/7v6gzYa4X6