FOR IMMEDIATE RELEASE: 17/Jul/2022
Welcome back to The Memo.
These emails now come from a new platform with better access to archives, simpler handling of credit card expiration and renewal, cleaner layout, faster publishing, and maybe even video backups.
The BIG Stuff
New article: Use cases for large language models including GPT-3 (15/Jul/2022)
Here are some of the major achievements and use cases of large language models (LLMs) like GPT-3. This list is not exhaustive, there are many, many use cases for large language models in 2022 and beyond. As the creator of LLM Aleph Alpha Luminous says:
“[LLMs can] Solve every problem that can be expressed with language.”
Read the article: https://lifearchitect.ai/use-cases/
Watch the video:
Meta AI releases NLLB (Jul/2022)
Meta AI has released NLLB (no language left behind), a 54.5B MoE model and a 3.3B dense model, covering 200+ languages.
“Driven by the goal of eradicating language barriers on a global scale, machine translation has solidified itself as a key focus of artificial intelligence research today. However, such efforts have coalesced around a small subset of languages [like English, Spanish, and French], leaving behind the vast majority of mostly low-resource languages. What does it take to break the 200 language barrier while ensuring safe, high quality results, all while keeping ethical considerations in mind? In No Language Left Behind, we took on this challenge by first contextualizing the need for low-resource language translation support through exploratory interviews with native speakers. Then, we created datasets and models aimed at narrowing the performance gap between low and high-resource languages.”
Announce: https://ai.facebook.com/research/no-language-left-behind/
And the beautiful demo: https://nllb.metademolab.com/
The original paper is 190 pages long(!), and includes a table of all languages used in training. Although, just as a personal side note, Meta AI seems to be missing the 363 Indigenous Australian languages (map below shows just a subset of these).
The Interesting Stuff
Exclusive first look: GPT-3 + DALL-E 2 + VLM (17/Jul/2022)
I think this is a world first, but I don’t mind being corrected! In this special edition, Leta AI writes a children’s book, and then its prompts are used to generate AI images via DALL-E 2. Lastly, we use Quickchat.ai Emerson and its visual language model to look at the images and comment on them. I had a lot of fun with this, and as a paid reader of The Memo, you get to see it first AND you get a complimentary copy of the eBook!
Watch the video:
Download the book:
NVIDIA matches Google in using AI to design its AI hardware (8/Jul/2022)
In my AI report last year (2021), I covered Google’s use of AI to design its current generation of TPU chips. TPUs are then used to train AI models.
https://lifearchitect.ai/the-sky-is-on-fire/
In 2022, NVIDIA has followed suit for its Hopper chips, which are up to 6x faster than the A100 chips used to train current AI models in 2021 and 2022:
“We demonstrate that not only can AI learn to design these circuits from scratch, but AI-designed circuits are also smaller and faster than those designed by [humans and even] state-of-the-art electronic design automation (EDA) tools. The latest NVIDIA Hopper GPU architecture has nearly 13,000 instances of AI-designed circuits.”
https://developer.nvidia.com/blog/designing-arithmetic-circuits-with-deep-reinforcement-learning/
Google AI is going further with robots + LLMs (Jul/2022)
Google is doing some fascinating stuff with robotics, now integrating GPT-3 into robots that can navigate the kitchen, as well as on the streets! There are actually two different teams:
Kitchen: Google Robotics (12/Jul/2022)
Lab: https://innermonologue.github.io/
Paper: https://arxiv.org/abs/2207.056...
Streets: Google Robotics + UC Berkeley + University of Warsaw (10/Jul/2022)
Lab: https://sites.google.com/view/...
Paper: https://arxiv.org/abs/2207.044...
Meta AI is researching the use of LLMs to improve Wikipedia articles
With a model they called SIDE trained on a dataset called SPHERE (2021, uses Common Crawl), Meta AI plans to make it possible to verify all citations on Wikipedia.
Read the paper (16 pages): https://openreview.net/pdf?id=qfTqRtkDbWZ
Comparing text-to-image platforms: DALL-E 2, DALL-E Mini, etc (12/Jul/2022)
This is a 3,000-word article by Kazimierz Rajnerowicz, comparing some of the publicly available text-to-image platforms. It’s a great read and look!
https://www.tidio.com/blog/generative-ai-dall-e-examples/
MIT has a new model for molecule drug discovery (12/Jul/2022)
MIT researchers have developed a geometric deep-learning model called EquiBind that is 1,200 times faster than one of the fastest existing computational molecular docking models, used in drug discovery.
AI in sport
AI is featuring heavily in the big sporting tournaments this year.
For tennis in Wimbledon, IBM Watson was engaged to predict winners:
https://thenextweb.com/news/inside-wimbledon-ibm-ai-engagement-tennis-fans
For football/soccer in the World Cup, AI is augmenting human referees:
https://www.theverge.com/2022/7/5/23195009/fifa-ai-automated-offside-ball-body-tracking-2022-qatar-world-cup
Statement on AI sentience
As I deliberately refuse to comment on AI sentience in the media (no, AI is not sentient), please accept this well-written article as a full statement and conversation. It cites Blaise Aguera y Arcas from Google Research, who was one of the team members reviewing the recent media distractions.
“Instead of rehashing debates about whether machines have souls, the ongoing double-helix relationship between AI & the philosophy of AI needs to construct more nuanced vocabularies of analysis, critique, & speculation…”
https://www.noemamag.com/the-model-is-the-message/
Toys to Play With
Speak to a virtual Ruth Bader Ginsburg (14/Jun/2022)
Oh, this one is fun! AI21 fine-tuned their Jurassic-1 178B model with Justice Ruth Bader Ginsburg’s 27 years of data, including Supreme Court Opinions (legal judgments), media interviews, and public speeches. The result is impressive. Speak with it for free…
You might like to try these examples (top questions from reddit.com/r/AskLaw):
Can the Secret Service testify against the president they were assigned to protect if they witnessed illegal activity?
Is what Equifax did considered criminal negligence?
Can an employer ask me to undress myself (or simply to lift my shirt or a pant leg) to check for any non visible tattoos?
Are you allowed to refuse to be a part of an HOA?
DALL-E 2 eBook (14/Jul/2022)
Guy Parsons put together this colorful 82-page eBook on DALL-E 2 creations, and how to get the most of the platform. Note that every single image in the book is generated by DALL-E 2 (even the little bits of clipart!), and they are in so many different styles. Thanks, Guy!
Download the 82-page DALL-E 2 PDF (21MB).
DALL-E Mini and DALL-E Mega (Jul/2022)
Besides the free, ad-supported Craiyon (formerly DALL-E Mini), there is another new free platform (sign-in required) to interface with these new DALL-E 2 clones (not affiliated with OpenAI). Play with them for free…
https://playgrounds.ai/models/dalle-mini
https://playgrounds.ai/models/dalle-mega
MidJourney Salt
Follow this crowd-sourced adventure by Salt, where you can vote on Twitter to help guide a choose-your-own-adventure story using MidJourney as the image or comic or story generator. Get involved for free…
Watch and play Salt: https://twitter.com/SALT_VERSE
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
It is incredibly challenging to keep up with the AI landscape right now, as new announcements are made daily. But you can count on me to provide you with up-to-date context and insights about the stuff that matters in the world of artificial intelligence, especially large language models…
All my very best,
Alan
LifeArchitect.ai