AI
Stability AI now has an outpatient tool called Uncrop.
Using anti GAN as a control
Runway Gen 2 has been released. Text to video race has begun truly.
MidJourney 6 is coming soon and there's lots of rumours on what it will do.
AI: Google
Google DeepMind plays videogames
Google products get more AI in them.
AI: Creations
Lord of the rings in anime
Giant inflated Skyscrapers
AI: Lists
GitHub has an excellent list of papers on AI
New AI Tools, a Touring test, and Vision Pro with machine learning
Shots you can take in Midjourney
Lighting styles in MidJourney
Apple uses AI, medical charts made by AI and more
Characters made in text to 3D AI and more
A database for problems and solutions with AI. It's looking for ethical problems though.
10 ways to make money using ChatGPT
AI: Opinions and Propaganda
You shouldn't have kids till AI is figured out
AI is capitalist and should be feared. The only thing to stop it is government control.
Fine Tuning AI
Adobe will cover any copyright lawsuits over AI creations made by its products.
Google Cloud AI can be used for healthcare.
Open AI being sued for AI accusing someone of a crime.
The future is terrible because AI isn't controlled by us.
Why being trained in the arts will help you with AI. Big thing is you shouldn't steal from other artists.
McKinsey has about half their employees using AI to do tasks. They have strict guidelines and demand permissions be given first. This is bureaucracy trying to control AI.
Realistic looking robot warns of AI destroying us all
A design firm made in a week
AI: Papers
SPQR has a new weight system
Analyzing video and audio for AI
Using two LLaMAs set for different things together to analyze videos.
Analyzing contextual bugs for hallucinations in compilations
Checking for deductive reasoning within the chain of thought in AI.
Code evaluation for LLMs
AI-PA is a story making AI. It creates interactive visuals.
ReWOO checks outside data and can be much smaller. At this point I think AI programs are becoming band names. I'm looking forward to Better than EZrA.
ChatGPT vs Open source LLMs
I just want to say I appreciate the effort you put into these posts, they don't get enough engagement for the amount of content.
Thank you. It has helped me with my Thesis project because now I have a long list of articles and news about AI.
Do you think that AI is a good or a bad thing overall after reading all that? Do you believe Elon, who said it would destroy humanity?
That's sort of the basis of my thesis project. I have seen a lot of back and forth but very little from anyone who actually understands it. So, my project was to design something using AI. If I had a question I made a smaller project to analyze the question. I haven't posted any of those projects here out of fear someone would trace it back to me.
The best opinion is that AI is a tool that helps you do things a large room of people could do. That large room represents several fields and talents. I can't program, but I can get ChatGPT to do something for me. I need some quick art, so MidJourney helps me. I need exact measurements and Revit does most of the work. Suddenly I am everything needed for a large design firm. I can make several products, and make it seem like I have several people doing the job.
This means those who don't understand AI are in trouble, and those who do can replace them. It's neither good nor bad. Imagine some guy with an office in his house that covers Architecture, Animation, videogames, and even books with ease.
We are in for a change, it will have a lot of difficulties, but it will transition to something cool.
AI is not intelligent. The end.
It doesn't understand shit. It's trained by human reinforcement (RLHF) to give results that sound "good". That random humans will say "Hey, that sounds intelligent / makes sense".
I think by now everyone heard of the case of a guy being blamed for something criminal (money laundering?) by chatgpt. He never did this.
Same with image generation AI. If I tell it "generate me a human" it doesn't know what a human is. I have to add negative prompts like "too many fingers" to even get to basic ideas like "a human has 5 fingers". You can't even say "human with 5 fingers", because it doesn't understand what a "finger" is.
That also is all the reason why "AI" in art works great. If it fucks up something like generating a shitty image - you just go again. If your AI doesn't detect a child running on the street - you have a dead child.
Fun recent article about the topic in general: https://lcamtuf.substack.com/p/llms-are-better-than-you-think-at