The Memo - 18/Jan/2023
AI21 Spices, GPT-4 delayed, AI passing the bar/CPA/medical exams, new lawsuits, and much more!
FOR IMMEDIATE RELEASE: 18/Jan/2023
Welcome back to The Memo.
In the Toys to play with section, we look at a new game ‘AI or human art?’, a short movie by Midjourney, a free kids storybook writer leveraging GPT-3.5 + DALL-E 2, and more…
The BIG Stuff
DeepMind (and Google) may lead the closing of published AI research (12/Jan/2023)
Demis Hassabis [CEO of DeepMind, part of Alphabet/Google]: “We’re getting into an era where we have to start thinking about the freeloaders, or people who are reading but not contributing to that information base,” he says. “And that includes nation states as well.” He declines to name which states he means—“it’s pretty obvious, who you might think”—but he suggests that the AI industry’s culture of publishing its findings openly may soon need to end.
Without wanting to get political, I have to comment on DeepMind’s suggestion that they no longer publish AI findings as a way of reducing IP theft or copying...
I became a permanent resident of China back in 2009, and a few years ago was based in a Russian-speaking country with strong economic ties to Russia. Both nation states have their own AI research labs, with varying degrees of output. Yandex and Sber in Russia, and mega-labs like BAAI and Tsinghua University in China continue to train new AI models and publish new AI papers. Members from all these labs are readers of The Memo.
The unique openness of the AI research field is a brilliant phenomenon. Right now, anyone can go to arxiv.org or github.com and read in-depth papers about up-to-the-minute artificial intelligence. Meta AI, a US-based lab, even released a blow-by-blow account of training their OPT-175B model (GitHub).
Without complete openness and strong collaboration between all AI labs, universities, governments, and other organizations, AI progress could come to a complete stop.
I would be disappointed and deeply concerned for humanity if Demis’ attitude prevailed. The guarding of AI research is one of the few things that could pause the rapid pace of change (and exponential expansion) we are experiencing right now.
Read the interview: https://time.com/6246119/demis-hassabis-deepmind-interview/
Read Google’s recent PR release related to this, with Demis’ sign-off: https://ai.google/our-focus/
Download the PR release as PDF for your own policy/organization reference: https://ai.google/static/documents/google-why-we-focus-on-ai.pdf
GPT-4 delayed, OpenAI training a text-to-video model (12/Jan/2023)
OpenAI CEO Sam Altman has ‘said OpenAI’s GPT-4 will launch only when they can do it safely & responsibly. “In general we are going to release technology much more slowly than people would like. We're going to sit on it for much longer…” Also confirmed [text-to-video] video model in the works…’
…Altman was asked if GPT-4 will come out in the first quarter or half of the year, as many expect. He responded by offering no certain timeframe. “It’ll come out at some point, when we are confident we can do it safely and responsibly,” he said.
When asked about one viral (and factually incorrect) chart that purportedly compares the number of parameters in GPT-3 (175 billion) to GPT-4 (100 trillion), Altman called it “complete bullshit.”
“The GPT-4 rumor mill is a ridiculous thing. I don’t know where it all comes from,” said the OpenAI CEO. “People are begging to be disappointed and they will be. The hype is just like… We don’t have an actual AGI and that’s sort of what’s expected of us.” – via The Verge (18/Jan/2023)
Read more: https://lifearchitect.ai/gpt-4
And that’s the end of the bad news…
The Interesting Stuff
Microsoft announces general availability of Azure OpenAI Service including APIs for GPT-3.5, Codex, DALL-E 2, ChatGPT (16/Jan/2023)
This is going to increase the number of enterprise customers with access to large language models. Here’s a look at OpenAI/Microsoft GPT-3 enterprise clients as of two months ago (Nov/2022):
In the press release, Microsoft revealed that KPMG and Al Jazeera are also notable customers as of 16/Jan/2023.
Read the release: https://azure.microsoft.com/en-us/blog/general-availability-of-azure-openai-service-expands-access-to-large-advanced-ai-models-with-added-enterprise-benefits/
AI21 Labs launches Spices within Wordtune (17/Jan/2023)
Since its release in Aug/2021, I’ve enjoyed Israeli AI lab AI21’s Jurassic-1 (J1) model. At 178B parameters it is larger than GPT-3. And I really enjoyed playing with this new writing tool functionality based on J1.
Using a choice of 12 cues, Spices works alongside a writer to assist in the writing process, generating a range of textual options to add to and enhance sentences. Spices can also suggest statistics to strengthen an argument or sharpen a detail.
A key feature of Spices is that it always attributes its sources, providing users with links back to the source, solving one of the major issues that many Large Language Models (LLMs) face today…
“Spices is a toolbox that melds the best that both man and machine can offer, working alongside writers as a source of inspiration for better, more efficient and more compelling writing, while ensuring that writers themselves have the space and freedom to best express their thoughts, insights and information.”
Try it yourself for free: https://app.wordtune.com/editor/
See a demo: https://www.wordtune.com/spices
Read the release: https://www.prnewswire.com/il/news-releases/with-launch-of-wordtune-spices-ai21-labs-lifts-the-curtain-on-the-future-of-writing-301723245.html
Read a related article: https://techmonitor.ai/technology/ai-and-automation/chatgpt-alternative-wordtune-spice
Read the (very dry) related paper referencing Retrieval-Augmented Language Models with GPT-2 only.
DeepMind Sparrow dialogue model may be released this year (12/Jan/2023)
Further to the Time interview above, DeepMind’s CEO did allude to beta testing Sparrow (based on Chinchilla) some time this year. Chinchilla is currently the largest dense AI model in the world based on training tokens (1.4T tokens trained to 70B parameters).
DeepMind is also considering releasing its own chatbot, called Sparrow, for a “private beta” some time in 2023. (The delay is in order for DeepMind to work on reinforcement learning-based features that ChatGPT lacks, like citing its sources. “It’s right to be cautious on that front,” Hassabis says.)
Read the interview: https://time.com/6246119/demis-hassabis-deepmind-interview/
Learn more about Sparrow via my documented rules: https://lifearchitect.ai/sparrow/
…and my video from Oct/2022:
Bill Gates on 2023 AI (11/Jan/2023)
Question: Hi Bill. Many years ago, I think around 2000, I heard you say something on TV like, “people are vastly overestimating what the internet will be like in 5 years, and vastly underestimating what it will be like in 10 years.” Is any mammoth technology shift at a similar stage right now? Any tech shift - not necessarily the Internet
Bill Gates: AI is the big one. I don't think Web3 was that big or that metaverse stuff alone was revolutionary but AI is quite revolutionary.
via Reddit.
GPT-3.5 passes the bar exam, CPA, & medical licensing exam (Dec-Jan/2023)
Bar: For best prompt and parameters, GPT-3.5 achieves a headline correct rate of 50.3% on a complete NCBE MBE practice exam, significantly in excess of the 25% baseline guessing rate, and performs at a passing rate for both Evidence and Torts.
Paper on passing the bar: https://arxiv.org/abs/2212.14402
CPA: …`text-davinci-003` appears to be approaching human-level performance on the Remembering & Understanding and Application skill levels in the Exam absent calculation. For best prompt and parameters, the model answers 57.6% of questions correctly, significantly better than the 25% guessing rate, and its top two answers are correct 82.1% of the time, indicating strong non-entailment.
Paper on passing the CPA: https://arxiv.org/abs/2301.04408
Medical: We evaluated the performance of a large language model called ChatGPT on the United States Medical Licensing Exam (USMLE), which consists of three exams… ChatGPT performed at or near the passing threshold for all three exams without any specialized training or reinforcement. Additionally, ChatGPT demonstrated a high level of concordance and insight in its explanations. These results suggest that large language models may have the potential to assist with medical education, and potentially, clinical decision-making.
Paper on passing the USMLE: https://www.medrxiv.org/content/10.1101/2022.12.19.22283643v1
Bonus: ChatGPT was listed as an author on this USMLE paper!
ChatGPT in VR (12/Jan/2023)
Are you sick of hearing about ChatGPT yet? What about if we put it inside VR?
GitHub litigation extended to SD + Midjourney (14/Jan/2023)
One commenter describes this misinformed litigation as ‘violently American’!
You may recall that the litigator—a typographer—won the October 2022 Who Moved My Cheese? AI Awards! for his work suing GitHub and Microsoft and OpenAI many months ago.
Read more about the new lawsuit: https://stablediffusionlitigation.com/
The complaint (PDF): https://stablediffusionlitigation.com/pdf/00201/1-1-stable-diffusion-complaint.pdf
Getty litigation against SD (17/Jan/2023)
This week Getty Images commenced legal proceedings in the High Court of Justice in London against Stability AI claiming Stability AI infringed intellectual property rights including copyright in content owned or represented by Getty Images. It is Getty Images’ position that Stability AI unlawfully copied and processed millions of images protected by copyright and the associated metadata owned or represented by Getty Images absent a license to benefit Stability AI’s commercial interests and to the detriment of the content creators.
Read: https://newsroom.gettyimages.com/en/getty-images/getty-images-statement
Article on The Verge: https://www.theverge.com/2023/1/17/23558516/ai-art-copyright-stable-diffusion-getty-images-lawsuit
Archisound: Text-to-audio generation via diffusion (17/Jan/2023)
Listen to the samples: https://flavioschneider.notion.site/flavioschneider/Audio-Generation-with-Diffusion-c4f29f39048d4f03a23da13078a44cdb
Read the paper: https://raw.githubusercontent.com/flavioschneider/master-thesis/main/audio_diffusion_thesis.pdf
Read the Twitter announce: https://twitter.com/flavioschneide/status/1615323264372326400
Toys to Play With
20-min Midjourney movie (16/Jan/2023)
A viewer from BaldTV made a 20-minute movie about the ancient underground city of Derinkuyu (Turkey). All images were generated via Midjourney, and the text/audio content is in Polish; YouTube shows English subs.
GPT-3.5 + WolframAlpha (16/Jan/2023)
Thanks to James Weaver for this excellent combination of GPT-3.5 (a large language model) and Wolfram Alpha (an answer engine)!
GPT-3 + Google Sheets in Japanese (16/Jan/2023)
Speak Japanese? Here’s a translation of my GPT-3 + Google Sheets functionality:
Japanese: https://liquidjumper.com/google-sheets/googlespreadsheet_connect_to_gpt_3
Original (English): https://lifearchitect.ai/sheets/
AI story generator (17/Jan/2023)
This platform uses GPT-3.5 and DALL-E 2, and is an alternative to the OnceUponABot.com platform used in my recent keynote in Brussels.
Try it (bring your own key): https://storiesgenerator.com/
AI art or human art?
This is a fun guessing game, and no login needed. The zoom level is whack, so I zoomed out to 25% in Chrome using Command -.
Try it: http://aiorart.com/
Ping me with a screenshot if you get a perfect score!
Next
With the delay of GPT-4, I’m looking at other more interesting developments happening this quarter. Count on it being a noisy one!
All my very best,
Alan
LifeArchitect.ai
Archives | Unsubscribe new account | Unsubscribe old account (before Aug/2022)