Hey everyone, welcome to another installment of What’s New in AI, a weekly newsletter summarizing the trends happening in the world of artificial intelligence. Lots of news this week, from GPT tutorials, to more OpenAI announcements, to lawsuits. Let’s dive in:
Building GPT from Scratch
Andrej Karpathy, independent researcher and previous Director of AI at Tesla, released a 2 hour video walkthrough of building and training a GPT model.
We build and train a Transformer following the "Attention Is All You Need" paper in the language modeling setting and end up with the core of nanoGPT.
If you’re building with GPT or just curious about how GPT works under the hood, this is an important video to watch and learn from.
Azure OpenAI Service
Microsoft has continued to make headlines with their OpenAI integrations. They just announced that their Azure OpenAI Service is now generally available, giving businesses access to advanced AI models such as GPT, Codex, and DALL•E 2. The service exposes the underlying technology that Microsoft has been using for their own products such as GitHub Copilot, Power BI, and Microsoft Designer.
Microsoft is making key plays to compete with Google Cloud and AWS. If you’re considering building and scaling projects with AI, it may be worth looking into the Azure cloud option.
OpenAI ChatGPT API
OpenAI announced a waitlist for ChatGPT API access. This is significant because anyone currently building products with GPT is actually using older models (GPT3 and below) rather than ChatGPT. Once the API access is released, it will provide developers with an even better model to work with, allowing for more natural and accurate language generation.
If you’re building with GPT, make sure to sign up for the waitlist here: https://share.hsforms.com/1u4goaXwDRKC9-x9IvKno0A4sk30
Stability AI Lawsuit
A group of artists have filed a class-action lawsuit against Stability AI, DeviantArt, and Midjourney for their use of Stable Diffusion (the generative image AI model).
The lawsuit focuses on the fact that the tool used unauthorized copies of billions of copyrighted images for training, which were made without the knowledge or consent of the artists. Since Stability AI is an open source company, they were transparent about their process for training Stable Diffusion, making the case for the lawsuit even easier.
Even assuming nominal damages of $1 per image, the value of this misappropriation would be roughly $5 billion.
(Source)
AI models will inevitably require more training data as they evolve. Legal battles such as this one will answer the question of responsibility for the use of unauthorized copyrighted material in the data for training.
Fun Bites
On the hardware side of AI, Boston Dynamics released a new video of their robot helping a construction worker by bringing tools up a scaffolding. Yet another peek into the not-so-distant future of how AI + robotics will revolutionize the way we work.
@Amir_dbab pulled off an awesome marketing stunt using generative AI to hack the TikTok algorithm. Check out his thread on creating an AI generated cartoon reporter talking about the latest news.
Generative AI is quickly hitting mainstream as more brands adopt it in their marketing and content creation. McDonalds just released a TV commercial using the NeRF (generative 3D) AI model.
@ammaar released an animated Batman movie, created solely through generative AI. It showcases how revolutionary the tech is for the world of animation.
Researchers at Cornell University developed an AI model that detects people’s position in rooms through WiFi signal…super Big Brother energy here.
That’s all the highlights for this week. Check out the Twitter list of AI builders who I’m following.
I'm hosting a What’s New in AI Twitter space every Friday at 12:30pm EST. Join to discuss the featured newsletter topics along with anything that didn’t make it in this week.
We also have a discord community! If you want to participate in conversations relating the latest developments in AI, join here: https://discord.gg/kApBrN8uqv