FOR IMMEDIATE RELEASE: 5/Dec/2022
Welcome back to The Memo.
I recently interviewed with Cris Sheridan from FinancialSense.com, about AI through to December 2022. Great questions from a network focused on business and finance.
Listen here:
I also released my end-of-year report to paid readers last week, and there is a video version available here:
The BIG Stuff
OpenAI releases new engine: text-davinci-003 (28/Nov/2022)
OpenAI quietly released their latest engine, the new GPT-3 default. It builds on InstructGPT, and is called text-davinci-003.
The only guidance they gave are these three points:
It produces higher quality writing. This will help your applications deliver clearer, more engaging, and more compelling content.
It can handle more complex instructions, meaning you can get even more creative with how you make use of its capabilities now.
It’s better at longer form content generation, allowing you to take on tasks that would have previously been too difficult to achieve.
Here’s a nice summary of what it can and can’t do: https://threadreaderapp.com/thread/1597374826305318912.html
I went live within minutes of its release, here’s the long video (1 hour 23 mins) where you can play along with a few hundred guests:
The Interesting Stuff
OpenAI releases ChatGPT (30/Nov/2022)
OpenAI’s ChatGPT is a hamstrung and ultra-safe fine-tuned version of the InstructGPT model, that is next to useless. The rules and policies seem to be very similar to those used by DeepMind Sparrow (LifeArchitect.ai).
ChatGPT gained 1 million users in just 5 days, according to the President of OpenAI!
I also went live with a demo of this one on release: Watch ChatGPT in action with a few hundred guests on YouTube.
ChatGPT is flooding Stack Overflow/Stack Exchange with AI data (Dec/2022)
These 3 threads will be in the history books. Interestingly, none of the human responses come close to satisfactorily answering this kind of query around AI, ethics, learning, education, knowledge.
Stack Overflow/Stack Exchange is a Q&A community site, founded by Joel Spolsky, and has been in the top 100 most visited sites in the world for about a decade. It also came in at item #20 for DeepMind’s Chinchilla training dataset, MassiveWeb, but it was not explicitly noted for GPT-3’s dataset. You can read more about that in my paper, What’s in my AI?. It’s a strange twist to see one of the largest AI training datasets now being used by AI to inject new content into that data source!
Together releases GPT-JT (30/Nov/2022)
The AI hardware lab, Together, have released a new version of EleutherAI’s GPT-J 6B based on Google Research’s UL2 and Chain-of-Thought (CoT) techniques. GPT-JT has been fine-tuned on an additional 3.53B tokens, and outperforms most other models on the HELM/RAFT benchmark.
Read more: https://www.together.xyz/blog/releasing-v1-of-gpt-jt-powered-by-open-source-ai
Play with the demo on HF: https://huggingface.co/spaces/togethercomputer/GPT-JT
AI art at MOMA in NYC (Nov/2022)
Watch Anadol’s video: https://refikanadol.com/works/unsupervised-machine-hallucinations-moma/
Read MOMA’s article: https://www.moma.org/magazine/articles/658
GPT-3 for talking with your younger self (Nov/2022)
Michelle Huang fine-tuned GPT-3 with her own childhood journal entries.
I trained an AI chatbot on my childhood journal entries - so that I could engage in real-time dialogue with my 'inner child'…
I kept diaries for about 10+ years of my life, writing almost everyday — about my dreams, fears, secrets... I used GPT-3 as my playground, and ended up taking samples of text from a bunch of different entries that I felt were representative of my personality and values during that time... this way, I could accurately simulate what it would be like to talk to my childhood self, based on real data sources during that time period.
I would say, for me personally - the kindness and understanding I felt from my 'past self' helped me untangle some knots and stuckness that I felt. It illuminated to me how hard I was on myself, but how much I extended understanding to others. It felt really rewarding to receive this kindness as well, from a younger version of myself.
From a broader learnings [perspective], what stood out to me was the ability for technology to be used for mental health and wonder.
Read the article: https://nwn.blogs.com/nwn/2022/11/michelle-huang-ai-art-gpt-diary-conversation.html
Disney announces re-aging AI (30/Nov/2022)
Disney researchers have created a new neural network that can alter the visual age of actors in TV or film. Face Re-aging Network (FRAN) is "the first practical, fully automatic, and production-ready method for re-aging faces in video images."
Read the paper: https://studios.disneyresearch.com/2022/11/30/production-ready-face-re-aging-for-visual-effects/
Read an article on Ars: https://arstechnica.com/information-technology/2022/11/disney-creates-ai-fountain-of-youth-for-actors/
Stable Diffusion generates 250 different pages of a 1987 Radio Shack catalog (1/Dec/2022)
View the thread in Mastodon: https://tilde.zone/@ftrain/109436259129597431
Toys to Play With
ChatGPT in your mac toolbar (2/Dec/2022)
Download: https://github.com/vincelwt/chatgpt-mac
Vale of Valor game concept marketing generated by ChatGPT + Midjourney + DALL-E 2 in 5 hours (5/Dec/2022)
This game concept marketing site took a reddit user 5 hours to make; it is a page only and the game does not exist. (Click the green ‘Play on Steam’ button at the top-right for info).
Check it out: https://valeofvalor.com/
ChatGPT generated this Python repo on Github (5/Dec/2022)
Take a look: https://github.com/vrescobar/chatGPT-python-elm
Write a book with GPT-3 (Nov/2022)
Thanks to Mateusz Drozd for this guide on how to write a book using OpenAI’s GPT-3.
https://drozd.gumroad.com/l/WriteABookUsingGPT-3
Next
What a spectacular start to the final month of 2022! I’m sure there are one or two labs with models up their sleeves for the last few days before 2023…
All my very best,
Alan
LifeArchitect.ai
Housekeeping…
Unsubscribe:
Older subscriptions before 17/Jul/2022, please use the older interface or just reply to this email and we’ll stop your payments and take you off the list!
Newer subscriptions from 17/Jul/2022, please use Substack as usual.
Note that the subscription fee for new subscribers will increase from 1/Jan/2023. If you’re a current subscriber, you’ll always be on your old/original rate while you’re subbed.
Gift a subscription to a friend or colleague for the holiday season: