GSV's AI News & Updates (03/20/22)
GPT-4, Google's AI Workspace and Microsoft's 365 Copilot, OpenAI Partners with Duolingo and Khan Academy, Adept's $350M Series B
Huuuuge week for AI (with GPT-4, Anthropic releases Claude, Google’s PaLM API, Google adds GenAI to workspaces, and Adept’s $350M funding all announced on the same day - Pi Day 3/14)!
You are receiving this weekly update on the intersection of AI and education as part of our GSV community – we hope this will be a helpful resource as we navigate this evolving and exciting intersection. Please feel free to send over any relevant resources you may come across (claire@gsv.com) – we’d love to learn and hear from you!
General 🚀
The first AI model that can “see”, GPT-4 can describe, analyze and perform tasks from images. “Napkin sketch to website” demo below:
or help you come up with recipes out of items in your fridge!
Superior reasoning skills and ability to handle complex tasks. It’s more intelligent than GPT-3.5, performing in the 90th percentile across multiple exams (the Bar Exam: 90%, LSAT: 88%, GRE Quantitative: 80%, a 1410 on the SAT)
You can control its style - while ChatGPT writes like a robot, GPT-4 introduces “steerability” to steer the model in style or direction you prefer
Can handle 25,000+ words (GPT 3.5 could only process 3,000)
Currently available in ChatGPT for premium users
Since the launch on last Tuesday, a handful of companies have disclosed that they had early access and have been building using GPT-4 - Intercom, Bing, Khan Academy, Duolingo and more
Generative AI is coming to Google Workspace: a massive drop for Google’s 3B+ users. New features include:
Summarize, brainstorm, proofread, paraphrase in Gmail and Docs
Auto-generated images, audio, and video in Slides
Go from raw data to insights and analysis via auto completion, formula generation, and contextual categorization in Sheets
Generate new backgrounds and capture notes in Meet and enable workflows for getting things done in Chat
Microsoft also launched 365 Copilot, “your copilot for work”: essentially ChatGPT for all Microsoft 365 apps and equips anyone using Word, Excel, PowerPoint, Outlook, Teams and more with their own personalized AI assistant.
Zapier launches Natural Language Actions: put Zapier's 20k+ app actions inside your product and use simple natural language to move info between apps.
AssemblyAI releases Conformer-1 speech recognition model: it’s achieving near-human level performance after being trained on 650K hours of audio data. It has 43% fewer errors on noisy data than other publications, a 29% speed-up at inference, and actually uses novel tech.
Putting GPT-4 to the Test - Can it code a complete game from scratch?
Baidu Plunges After Recorded Demo of AI Chatbot Disappoints: Baidu rushed to get its ChatGPT competitor Ernie launched, and the cut corners showed. Its stock fell as much as 10% on the news.
OpenAI CEO Sam Altman and CTO Mira Murati’s Interview with ABC: the OpenAI team does an excellent job acknowledging AI’s challenges and taking on some tough questions, but also highlighting AI’s potential to revolutionize industries and improve lives."It is going to eliminate a lot of current jobs, that's true. We can make much better ones."
Reid Hoffman wrote a book called Impromptu with Open AI’s GPT-4
Education and the Future of Work 📚
Khan Academy announced GPT-4 powered Khanmigo: an AI powered assistant that functions as both a virtual tutor for studetns and a classroom assistant for teachers. One of GPT-4’s chief capabilities is being able to understand freeform questions and prompts. That ability—to have a human-like back and forth—provides Khan Academy with perhaps the most key capability: asking each student individualized questions to prompt deeper learning.
Duolingo launches Duolingo Max, a learning experience powered by GPT-4: a new subscription tier above Super Duolingo that gives learners access to two brand-new features and exercises: Explain My Answer and Roleplay. Both are powered by GPT-4.
Ethan Mollick: Using AI to make teaching easier & more impactful: five strategies and prompts that work for GPT-3.5 & GPT-4.
Paul LeBlanc: We’re Asking the Wrong Questions About AI: It’s not only about academic integrity: higher ed faces much bigger challenges in the face of frightening questions about the future of the knowledge workforce.
AI Is About to Transform Childhood. Are We Ready?: In the future, middle-class kids will learn from, play with and grow attached to their own personalized AI chatbots.
LinkedIn is adding AI tools for generating profile copy and job descriptions
PwC announces strategic alliance with Harvey, AI lawyer copilot: to support 4,000 of its lawyers in over 100 countries.
Startups and Tools 🛠️
Adept raises $350M in Series B funding co-led by General Catalyst and Spark Capital: Adept is building on this momentum via a new kind of foundation model that can perform actions on any software tool using natural language. Adept’s vision, at a high level, is to create what it refers to as an “AI teammate” trained to use a wide variety of different software tools and APIs - allowing AI to work with you on your computer, letting it develop with you and execute operations for you at your request.
Stability AI launches Stable Diffusion Reimagine: Clipdrop tool that allows users to generate multiple variations of a single image without limits.
Bloop: understand codebases with GPT-4 and semantic code search.
WhatDoesThisCodeDo: uses chatGPT under the hood to explain any piece of code you don’t understand.
Botz: set your blog on autopilot with AI - SEO-optimized articles with complete automation.
Magic Slides: AI slides and presentation generator.
Storywizard.ai: create personalized and engaging children’s stories using AI.
Tech 💻
Google opened access to PaLM’s API: developers can now build and customize their own content and applications on top of the LLM. The API also comes with an intuitive tool called MakerSuite, which lets you quickly prototype ideas and, over time, will have features for prompt engineering, synthetic data generation and custom-model tuning.
First Open Source Text-to-Video Model Released: the model, available on Hugging Face, is trained on 1.7B parameters. Example of someone making their own Star Wars clip using AI. (Note: because it is trained on shutterstock content, most outputs are watermarked).
ViperGPT: A Python Framework that Combines Vision and Language Models Using Code Generation to Achieve State-of-the-Art Results
Fun/Other 🎉
Strangeworks uses only Generative AI to deliver keynote at SXSW: everything from the slides and graphics for the presentation to the publication handed out to attendees to the final script.
ChefGPT: AI powered digital chef with recipe recommendations, meal plans creations and more.
The Infinite Arcade: GPT-4 to code any arcade game you can think of and let you play it instantly.