The Memo - 13/Aug/2022
ARK predicts $24 trillion value in AI models, media begins using AI art, new robots, and much more!
FOR IMMEDIATE RELEASE: 13/Aug/2022
Welcome back to The Memo.
I recently spoke with the largest university-based research site dedicated to machine learning in Australia, The Australian Institute for Machine Learning (AIML). These guys are part of The University of Adelaide, and famously beat Google, Intel, Oxford, IBM, and Carnegie Melon in ImageNet parsing.
This discussion has some great questions, and I had a lot of fun with interviewer Tim Whiffen:
Or you can listen via Stitcher or Apple Podcasts or Spotify.
The BIG Stuff
I know we have a few readers of The Memo who are policy advisers for government and IGO. If you work in business and enterprise, this should also be interesting. If you are a human being, this should be interesting, too!
Recall this quote from Masayoshi Son, founder, chairman and CEO of the multi-billion dollar Japanese holding company SoftBank:
It may look like we are investing [$100 billion through SoftBank] on a whim without any consistency, but one common theme is artificial intelligence… I have shifted entirely so that I am devoting 97% of my time and brain on AI.
— (Forbes blog with citations)
ARK Invest has been particularly vocal in the last few days, and their reports are always excellent analyses (alongside McKinsey) on the state of AI and forecasting.
ARK Invest: Neural net capability should increase by ~5,000x by 2025 (13/Aug/2022)
Get ready for further explosions in the next few months…
Neural net capability should increase by ~5,000x by 2025… What might [this neural net capability/Transformer model/LLM capability increase of ~5,000x by 2025] tangibly mean? If the complexity of generating algorithmic video scaled linearly by number of pixels generated (which there are puts and takes), then 2.5 minute HD videos of DallE-2-level capability could be expected in the 2025 timeframe.
A related chart—again by ARK Invest—shows AI value measured in the trillions of dollars within the next few years, with LLMs alone (‘Foundation AI Model-as-a-Service APIs’ like GPT-3 or DALL-E 2) forecast to generate business value of around $24 trillion (Twitter source, article source, slide deck source pp22):
The Interesting Stuff
New model viz (8/Aug/2022)
and a more complete list at: https://lifearchitect.ai/timeline/
Roadmap video released (12/Aug/2022)
China launches new robots, before Elon Musk can launch the Tesla Optimus (Aug/2022)
Xiaomi CyberOne launch video (12/Aug/2022):
Xiaomi CyberOne on-stage demo (12/Aug/2022):
Tencent Max (8/Aug/2022):
Robotics X Lab pays more attention to the research of robot’s autonomous characteristics, aimed at realizing the robot’s autonomous judgment, decision-making and task completion in dynamic environments with great uncertainty.
Read the article (8/Aug/2022).
Boston Dynamics splits out a new AI lab with half a billion dollars funding (13/Aug/2022)
Boston Dynamics AI Institute (BDAII) will spearhead advancements in artificial intelligence and robotics.
To achieve such advances, the Institute will invest resources across the technical areas of cognitive AI, athletic AI, and organic hardware design, with each discipline contributing to progress in advanced machine capabilities. In addition to developing technology with its own staff, the Institute plans to partner with universities and corporate research labs.
Read the article: https://spectrum.ieee.org/boston-dynamics-ai-institute-hyundai
The Atlantic begins using AI text-to-image for stock photography (9/Aug/2022)
News outlet The Atlantic posted an article about Alex Jones, using MidJourney to generate the header image.
This is much bigger news than it may appear at first glance. Stock photography is being replaced right now, and it happened so fast!
Prompt: "Alex Jones inside an American Office under fluorescent lights"
AI art by Midjourney.
Pay for your groceries with your hand, in California with Amazon (9/Aug/2022)
Amazon’s palm-scanning technology is expanding to 65 Whole Foods locations across California… [including] Malibu, Montana Avenue, and Santa Monica… you can take items to checkout and… a hover of your hand over the device is all that’s needed to pay and leave.
Toys to Play With
Ask Alexa to write a children’s story with GPT-3 (Aug/2022)
If you like getting your hands a little dirty, here’s some components to hook up together. The most expensive part is Zapier, at ~$50/month. Great application of GPT-3 and a smart device!
https://www.seotraininglondon.org/alexa-gpt-3-child-story/
Play AI-generated art hangman (8/Aug/2022)
This is kinda fun. Guess the prompt used to generate the image by guessing one letter at a time, hangman style. After you’ve read the instructions, click the ‘X’ at the top-right to begin.
Play: https://daydreams.ai/
Read a tech-noir story by my friend in Germany + GPT-3 (2021)
I LOVED reading this story by a Leta AI viewer in Germany, using the massive GPT-3 language model. His pen name is Derek Beauregard. It's the length of a proper book, though unedited. This is an exclusive, he is not releasing this publicly, and there is nothing like this in the wild (or on Amazon) yet. Thank you so much to D for releasing this book to The Memo readers at no fee! Fall in love with Lucy and friends in Terminal City Chronicles (PDF).
Next
As soon as Tsinghua releases the paper with benchmark results, I plan to release a report card on China’s GLM-130B model (‘The world's best public, open source language model’). In the interim, you can browse the GitHub GLM-130B repo.
I wonder what’s next… I’ll continue keeping pace. See you soon!
All my very best,
Alan
LifeArchitect.ai