🔎 What’s good today:

• 💻A bunch of new Google AI features

• 🦾 Start Using AI on your big documents

• 💬 ChatGPT to understand anything

• 👀 AI will soon be able to experience the world as humans do

• 👩‍🏫 ChatGPT Is Incredibly Smart and Shockingly Stupid

🔥Google makes a strong come-back

Google's recent I/O event was an exciting showcase of AI-centered innovations that make it clear the tech giant is evolving as an "AI-first" company. Here's how it will affect you and your business:

• Duet AI for Google Workspace enhancements, including AI in Gmail, Docs, Sheets, and Slides

• Google Maps to offer an 'Immersive View for Routes'

• Google Photos is getting a 'Magic Editor' feature

• Google Search will become interactive, responding to questions in a conversational manner like ChatGPT

• PaLM 2, Google’s powerful new LLM that will power over 25 new Google products

• Bard upgrades: waitlist removed, available in 180 countries, moved to PaLM 2

• Partnership with Adobe Firefly, making image-to-text available directly in Bard

• Med-PaLM, A LLM from Google Research, designed for the medical domain

• MusicLM, describe a musical idea and hear it come to life

• StudioBot, an AI coding bot for developers

Some interesting facts from the event: Google mentioned "AI" 143 times in their presentation, and Google stock increased by $56 billion.

🤓Finally - LLMs that work on huge documents

Anthropic, the AI company, has just expanded the token context for their conversational AI, Claude, from 9K to 100K tokens! Which is huge, as OpenAI and Bard only allow up to 3k tokens.

For those who don't know, a token is a measure of how many words a model can consider and recall. To put this in perspective, Harry Potter is around 100K tokens long, which means Claude can now digest the entire book and respond to you in just 22 seconds!

This expanded context window is a big deal. Researchers can easily find key points in long, technical papers, and business analysts can upload an entire 10-K from a corporation and watch Claude break it down for them. But here's the catch: Anthropic is only releasing this as part of Claude's API service, so for now, only businesses and AI researchers will be able to play with it.

Token restrictions are now a game of haves and have-nots, and even OpenAI is gating access to their largest context window, GPT4_32K. It's exciting to see these companies make these competitive leaps, but it will be a while before they trickle down to public use.

👀 Bringing AI Closer to Experiencing the World Like Humans

Exciting news from Meta! The tech giant has just open-sourced an AI model called "Image Bind", which has the unique ability to link together six types of data into a single embedding space.

Unlike most generative AI tools that can only link together two types of data, like text and image, Image Bind can understand information from images, text, audio, depth, temperature, and movement. This capability is akin to how humans perceive a sensory experience from a single image.

While Image Bind is currently a research project with no immediate practical applications, it is a significant milestone towards creating AI models that can generate multisensory experiences like humans. This breakthrough in AI technology is yet another example of how Meta is leading the charge in innovation and pushing boundaries to shape the future.

👩‍🏫 AI Is Incredibly Smart and Shockingly Stupid

Computer scientist Yejin Choi great talk where she demystifies the current state of massive artificial intelligence systems like ChatGPT, highlighting three key problems with cutting-edge large language models (including some funny instances of them failing at basic commonsense reasoning.) She welcomes us into a new era in which AI is becoming almost like a new intellectual species -- and identifies the benefits of building smaller AI systems trained on human norms and values.

Hope you enjoyed today’s email, see you next week!
Tony