Welcome, humans.
Most people are in your Instagram DMs. We're in your inbox every day at 5am. We are not the same.
Here’s what’s going on in the world of AI today:
- 🌩 AI is coming to Adobe's Creative Cloud
- 🔥 Sam's hot take on AI models
- 🌝 Elon Musk announces TruthGPT
- 🔬 Chat with arxiv research papers
Adobe Firefly Brings AI To Video-Editing 🎨
The number of professional-quality TikToks is about to skyrocket.
Adobe is bringing its generative AI suite Firefly into its Creative Cloud, making video customization more accessible than ever.
Enter text-to-video-editing: with Firefly, users can edit videos with simple text prompts...
- Change a video's color grade, background, and shade.
- Create subtitles, logos, and custom graphics.
- Convert scripts into detailed storyboards, pre-visuals, and b-roll recommendations.
...all without lifting a finger. This is huge for businesses aiming to create high-quality video and audio content without hiring expensive specialists.
It's also huge for Adobe. Generative AI could offer a major advantage for its existing millions of creative users, supercharging its subscription business.
Still, it's an ambitious promise for a service that has yet to make it out of beta.
Unlike other platforms that collect art without discretion (cough cough Midjourney), Adobe trains Firefly on publicly accessible images or those within its Adobe Stock database. This could mean more limited image generation capabilities.
Bottom line: We'll need to wait until later this year to see Firefly's true capabilities. Either way, the creative world is set for a complete transformation in video production.
Bigger Models ≠ Better Performance, Says Sam Altman 🔌
Crypto bros on the GPT-4 hype train earlier this year made the image shown above, which claimed GPT-4 would have 100 trillion parameters.
This turned out to be too much hype, of course. It has about 1 trillion.
Earlier this week, OpenAI CEO Sam Altman said the next best AI models won't necessarily have more parameters.
Which means even if you did make a model that had 100 trillion parameters, it wouldn't necessarily be that much better than GPT-4.
The strategy that got us from GPT-2 to GPT-4 ("use the same architecture, just make it a lot bigger") is coming to an end. That explains OpenAI's statement that they're not currently training GPT-5.
So what will bring us to the next phase?
- Nick Frosst, the cofounder of Cohere, thinks that better AI models will have to originate from new model designs and/or fine-tuning existing models based on human feedback.
- Researchers are building new architectures that could enable us to 50x the amount of context. In non-technical speak: If it works, you could upload 1,000 pages of documents/code (aka all your company's documents) to a ChatGPT-like app!
Whatever it is, we're watching closely, and you'll hear about it here first.
Around the Horn 🦄
- A German artist refused to accept his prize for the Sony World Photography Award after announcing he used AI to create the photo.
- Elon Musk announced his initiative to create 'TruthGPT,' an AI that tries to understand the universe.
- Snap released a new feature that lets users chat with fictional and historical characters.
- Google CEO Sundar Pichai went on "60 minutes" to discuss AI's acceleration and disruption to society.
- Universal Music Group announced its displeasure with training generative AI on its artists' music (like Drake).
- Meta built DINOv2, a method for training high-performing, self-supervised computer vision models.
Treats To Try 🍿
- Featured: Build your own chatbot with ChatGPT Builder. Powered by the same model as ChatGPT. Layer on your own customizations in plain English. Start a free trial with ChatGPT Builder.
- Single Prompt AI is a collection of single-purpose AI tools from itinerary builders to decoders.
- Chat with an arxiv research paper in <45 lines of code (thread).
- Build beautiful landing pages in seconds with Uncody.
- Shorty integrates into Siri to let you dictate questions to ChatGPT.
- ThinkGPT is a Python library that helps GPT think, reason, and create generative agents.
DM us links on Twitter: @nonmayorpete & @noahedelman02.
Tuesday Ticker ⏰
Here are the results from last week's poll:
Editing prompts is the new editing text.
Here is today's poll:
If AI-generated photos are allowed in photography competitions in the future, how should they should be judged?
- 🎨 Based on their aesthetic value and adherence to the competition's criteria
- ✍️ Based on the level of creativity and skill involved in generating the photo using AI
- 🧑⚖️ Based on a separate category or classification (comment what!)
Login or Subscribe to participate in polls.
We're excited to hear what you think!
A Cat's Commentary 😻
That's all we have for today!
- Enter the giveaway for ChatGPT Plus by referring your friends to The Neuron by 4/21!
- See you cool cats on Twitter if you're there: @nonmayorpete & @noahedelman02.
- Get your brand in front of 55,000+ professionals by sponsoring The Neuron.
- Was this email forwarded to you? Sign up here.
