Hi folks!👋🏻 This is The Prompt! & we are here to bring the latest in AI, even if we’re fashionably late sometimes (like today) 🤭

Let’s get it

FEATURED

JARVIS - ChatGPT to rule them all

It seems that Chat is becoming the go-to interface for AI, all because of ChatGPT.

But what if we could use ChatGPT not just for chat, but also as a way to control and execute different AI models?

Hello JARVIS! 👋🏻

This project is one of the most trending on GitHub and is setting the direction of AI for the next few months: Collaboration!

With JARVIS you can manage and execute different tasks from various AI models, via ChatGPT.

Interested to learn more?

Read below.

Here’s a simple example for you:

You: “Hey JARVIS, please generate a photo of a girl reading a book, with the same pose as the boy in this picture. Then please describe the image with your voice”.

Jarvis will then call a bunch of different models that will:

• analyze and extract the pose (model: OpenPose)

• extract the object location of the image (model: DETR)

• classify the image (model: ViT)

• extract the caption of the photo (model: ViT gpt2)

• generate the voice description (model: FastSpeech2)

Here’re those same steps visualized:

Powerful, right?

And this is just one vanilla example — imagine what else you can build by connecting ChatGPT with all these open-sourced models that are accessible with just 1 click on Hugging Face.

NEW TECH

Segment Anything - by Meta

The Segment Anything Model (SAM) by Meta can “cut out” any object from an image with a single click.

This is probably the "GPT-3 moment" for image segmentation.

This model can guess and outline an unfamiliar object without any previous training.

And Meta has been doing some homework because both the model and data (11M images, 1B masks) are open-sourced!

Try the demo here.

WHAT ELSE IS GOING ON

🙊 We can now build apps with our voice. In a demo posted on Twitter, Mckay Wrigley showed that his GPT-4 assistant can successfully create a project, build an app, GitHub repo, and deploy it to Vercel. All with his voice.

👀 Expedia adds conversational trip planning powered by ChatGPT. You can get travel recommendations, data-driven price tracking, & personalized travel options.

🏋🏻‍♀️  Google launched their newest AI supercomputer - they claim it beats Nvidia. They say that the supercomputer is “1.2x–1.7x faster and uses 1.3x–1.9x less power than the Nvidia A100”.

RESOURCES

The best resources we came across lately that will help you become better at writing prompts & building AI apps.

📚 Train Llama with RLHF on preference data [End-to-end tutorial ]

👋🏻 What it feels like to work in AI right now [ Opinion piece ]

🎥 Generate character videos from pose and text description [ Free Demo ]

TOOLBOX

The latest AI tools to use or get inspiration from.

• LayerUp - ChatGPT for data analytics

• Booom - AI generated game

• Article-Audio - Convert articles to human-like audio

• Minimap - Game recommendation chatbot

• June - Answer product questions using AI

• Mayday - AI powered calendar

• Speak Aide - Generate video scripts with AI

PROMPT OF THE DAY

TOOL

Midjourney

PROMPT

Cinematic, 35mm lens, f/1.8 Fashion Photography 🐯 🍓 , Flirty instagram african influencer, fearful expression, full body shot inspired by WLOP + Sakimichan, wearing a short fit pink latex ultraminiskirt and top, blonde hair, balloons, dreamy, candy colors --ar 16:9 --s 1000 --v 5

RESULT

LATEST PAPERS

• Instruction Tuning with GPT-4

• Diffusion Models as Masked Autoencoders

• Zero-shot Generative Model Adaptation via Image-specific Prompt Learning

• DreamFace: Progressive Generation of Animatable 3D Faces under Text Guidance

• SegGPT: Segmenting Everything In Context