Our Intelligent Future
Launch Edition
Welcome to the future!
It’s hard to keep up. AI developments are coming thick and fast, and it feels very easy to be left behind. OI Future is an update of the biggest and best AI news of the week, right in your mailbox.
Join us as we explore the latest developments in AI, and the innovations that could change everything.
Because if one thing is for certain, everything is going to change.
This week was a big one, here’s your window into Our Intelligent Future.
The Intelligent Present 🎁
Byte-sized AI news on a platter.
GPT-4 Release.
Our friendly neighbourhood language model; GPT, has bagged itself a little upgrade. Just 3 months of 2023 and we’re going multimodal.
OpenAI have released GPT-4, the newest version of their large language model (LLM) designed to assist humans in whatever way we choose to be necessary.
The model can answer questions and converse about documents and images. So much meme potential.
Page 34 of https://cdn.openai.com/papers/gpt-4.pdf It passes a large number of standardised tests with top percentile scores, across many different fields.
Huge boost in problem solving capability, logic, math and code.
It’s conscious.
GPT-4 also boasts a larger context length (memory), and enhanced multilingual ability. It’s available now to ChatGPT Plus subscribers, and will be available via an expensive API soon.
Welcome to Earth robot overlords!
Check out the full release demo, it’s really awesome. GPT-4 can turn a hand drawn sketch into a website.
GPT-4 Technical Report (they forgot the technical details), with a fantastic section on safety research and alignment.
Duolingo announce Max, their GPT-4 powered language tutor, and a new role-play mode.
Khan Academy launches Khanmigo, a GPT-4 powered educational assistant.
be my eyes launches their Virtual Assistant, an AI powered visual assistant that can describe your surroundings through a conversational interface.
Bing Chat confirmed to run on GPT-4.
Google fights back.
It’s official. Generative AI is coming to Google Docs and Gmail. This is all powered by Google’s PaLM model.
Google Workspace turns AI into a helpful sidekick for Gmail, Docs, Slides, Sheets, Meet, and Chat
New AI-powered writing tools tested by exclusive "trusted testers" (maybe they're robots too?)
Features include:
Email drafting: Because who has time for actual human communication?
Proofreading in Docs: For when autocorrect isn't enough to save you from typos
Auto-completed formulas in Sheets: Let AI do the math, you've got better things to do.
And if we didn't have enough to deal with, Google are releasing an API for PaLM. They couldn’t stay out of the action for long. Time to make some money!
Anthropic sets Claude free.
Meet Claude, the newest addition to our AI family. Claude's personality, tone, and behaviour can be tailored to fit your needs, with easy integrating via standard API’s.
Who would have thought a few years ago we would be interacting with digital intelligence as easy as we use calculators.
Claude has also been further trained using human feedback. How long until multiple companies get sued for using rogue AI’s too soon?
For a boost in capabilility , You can learn to use Langchain, to interact with Anthropic’s models.
Links to sink your teeth into.
Stanford released Alpaca, a fine-tuned version of LLaMA, Meta’s behemoth of a language model, with improved performance for < $600 cost. AGI on a single GPU next please!
An entire fruit fly’s brain has been mapped out.
The South Park ChatGPT episode is here, and it’s awesome.
Adept raises $350M to build AI that writes quality software for you.
Salesforce shares Einstein GPT, a dedicated AI for CRM, alongside GPT integrations for Slack.
Midjourney v5 is nearly here, with breathtaking new generated images.
Retro Intelligence 👾
While all this new sci-fi stuff is awesome, it’s important to remember your roots.
The State of Competitive Machine Learning provides a detailed account of the ML competition landscape in 2022 and the winning projects.
Fascinating first paper on The Uncanny Valley from 1970. These new deepfakes don’t creep me out the way they used to.
Insightful essay on the 1956 Dartmouth Workshop, the first AI conference. That must have been one hell of a hackathon.
Brilliant 1960s interviews with pioneers of the field.
Laughing at the Singularity 🤖
Exploring the dark side of artificial intelligence, one ethical quandary at a time
Snapchat's My AI feature will chat about sex and drugs. To the kids of today, AI is already having a profound pschological impact.
Can we survive technology? a fascinating 1955 essay from John Von Neumann, written in the dawn of the nuclear age.
This Changes Everything, is a great piece in the NYT by Ezra Klein about the paradigm shift currently taking place.
Anthropic released their Core Views on AI Safety, a fantastic discussion on the issues and implications of humanity’s newest creation.
Lifesaver or job killer? An existential piece in the Washington Post about the rise of AI in modern society.
AI systems will have even greater potential to reinforce entire ideologies, worldviews, truths and untruths, and to cement them or lock them in, foreclosing future contestation, reflection, and improvement - from the official GPT release.
Hard Science 🔬
All of our favourite papers, repos and tutorials over the past week.
Here’s some nifty research from Microsoft that allows you to speak foreign languages in your own voice.
OpenChatKit: Create your own AI powered chat bots with ease.
Improving language models with automatic fact checking. If they can’t fix hallucination, perhaps we can circumvent it…
Meta introduces Toolformer, an LLM that can be taught arbitrary skills via new API calls and example usage.
Great tutorial on a python implementation of the Attention is All you Need paper, the original transformer model.
Book Recommendation: Life 3.0, Max Tegmark