Runways Gen2 text-to-video is here, and it's mind blowing. Truly another 0-to-1 AI moment.
Here are some of the coolest examples of a new tool that just started rolling out in beta a couple hours ago 🧵👇
2/x For a bit of background @runwayml Gen1 video-to-video was already pretty amazing. Creators could use images and video to modify other videos and produce something entirely new:
4/x Over the last week or so, some folks with early access (who tended to be @runwayml employees) dropped some teasers that had the community salivating over the possibilities.
5/x Welp, earlier today they started rolling out the beta and the first prompts people are trying are already *so cool*. Here are some of my fav so far:
"a woman wearing a white dress meditating, standing on water, peaceful day, the most beautiful scene"
🥘Look at an image of food and spit out a recipe
🖼️Look at an image of art and tell you who made it and how it makes people feel
🌅Look at a sunset photo and write a poem
That AI is here and it's called MiniGPT-4.
If you want to learn more about MiniGPT-4, here are a few resources.
News of Google's Project Magi broke yesterday. It's a total reimagining of the search experience that has shaped our interaction with the internet for two decades. List of ten results OUT. Chatbot interface IN.
Amazon appears to be taking a developer-first approach to AI. Last week they announced Bedrock, which allows devs and companies to build off base models from @StabilityAI@AnthropicAI and others, and CodeWhisperer which is akin to Github Copilot
From Amazon getting in the game to AutoGPT enabling task lists that complete themselves to viral Drake songs that Drake didn't actually perform on, it was another big week in AI.
2/x 𝗚𝗲𝗻𝗲𝗿𝗮𝗹 𝗡𝗲𝘄𝘀
•Amazon Bedrock lets devs build on models like Stability AI and @AnthropicAI and starts previewing their Titan FM language model
•@StabilityAI releases latest update Stable Diffusion XL which improves photorealism and improves text in images
3/x 𝗚𝗲𝗻𝗲𝗿𝗮𝗹 𝗡𝗲𝘄𝘀 (cont.)
•Lots of chatter about Elon starting an AI company, incorporating X.ai and hiring AI engineers
Akira Sakamoto described it: "Starting from an initial task, Babyagi utilizes GPT-4 to generate solutions and new tasks, storing the solutions... for further retrieval."
One of the most compelling ideas I've heard lately is the idea that the Fed's forward guidance is effectively useless owing *not only* to questions of Fed credibility, but because of the broader breakdown in consensus reality.
In episode 299 of @HiddenForcesPod he and @JonAskonas put scholarly weight behind a feeling we can all recognize: that we no longer live in a world where there is one "truth," but a never-ending competition b/n competing truths
We see this in the realm of politics, of course, but also in the interpretations of events, news -- basically anything.
In the US we tend to ascribe it terms like "culture war" but it is actually something more fundamental.