Hi folks!👋🏻 This is The Prompt! And we have a packed newsletter today! Buckle up!

Here’s what we have for you today:

• Assistants API: Everything you need to know

• xAI PromptIDE

• Silent winner behind OpenAI RAG pipeline & xAI’s real-time retrieval

• OpenGPTs: open-sourced GPTs

• Multimodal AI: What can you build & models you can use

• Try the text-to-speech model by OpenAI

Let's read on

FEATURED

Assistants API: Everything you need to know

Huge announcements from OpenAI on Monday, but the biggest one was the Assistant API: an API that enables developers to build AI assistants within their own applications.

We cover everything you need to know:

What is it?

The Assistants API allows you to build AI assistants within your own applications. An Assistant has instructions, leverages extra knowledge, and can call models and tools to perform tasks.

To do this, this API uses Code Interpreter and Retrieval as well as function calling to handle the tasks that you’d usually do with external tools. For example, in some use-cases you won’t need to set up custom embeddings model, chunk up data or use a vector database for data retrieval; three valuable processes that power all of the the chatbots you see today.

You can try to build an assistant in the playground.

How much does it cost to retrieve data ?

OpenAI Assistant API costs $0.20/GB/day, which is more than what Pinecone charges ($0.1611/GB/day)

The argument here is that OpenAI does more than Pinecone, because it does the chunking/embedding for you, and Pinecone only vectorizes the data for you.

How many files can you host?

Currently you can only host a maximum of 20 files at 512MB each, so if you max it out you'll be storing 10GB at $2/day or $60/month for now. For some use-cases this can be very limiting.

Build without coding

The same capabilities that power the Assistant API, also power their new user-facing product: GPTs.

GPTs are are custom versions of ChatGPT like “apps” that will be listed in their “GPT store”, and everyone will be able to build them, no-code style. This is their second attempt of creating a marketplace of chatbots, or their own “App Store”. You’ll be able to list them, and share the revenue with OpenAI (still tbd how that will work)

What can you build?

You can build almost anything with this API: natural language-based data analysis app, a coding assistant, an AI-powered vacation planner, a voice-controlled DJ, a smart visual canvas—the list goes on.

🚨 What else is going on

• xAI announced the xAI PromptIDE which is an integrated development environment for prompt engineering and interpretability research.

• The silent winner behind all these announcements this week is Qdrant, which is the vector database behind OpenAI’s RAG system, and xAI’s real-time twitter data embedding

• New paper by Google provides evidence that transformers (GPT, etc) cannot generalize beyond their training data

• Langchain announced OpenGPTs - an open source GitHub repo enabling similar functionality to OpenAI's GPTs

📕 Resources

• ^[article]  Multimodal AI: What can you build, use-cases, models

• ^[video] OpenAI Assistants coding tutorial

• ^[thread] How to generate avatars with ChatGPT + Dalle3

• ^{[huggingface space]} Try the latest text-to-speech model by OpenAI

🧰 Tools

• Attio : Automate GTM strategies with AI

• BeeDone: Stay motivated with AI

• CommandBar: AI powered user assistance for mobile apps

• Codebay: Learn Python with AI tutor

• Mayday: The AI-Assisted Calendar

✍🏼 Prompt of the Day

TOOL

Midjourney

PROMPT

supermodel, posing, bright sunlight, outdoors, streetwear extravaganza, haze, Leica, halation, analog, 35mm ::1 prism light leaks ::2 --chaos 4 --ar 2:3 --style raw --v 5.2

RESULT