The Memo - 21/Mar/2024
Bill of Rights for AI, Grok-1 314B openly released, GPT-4 plays Doom, and much more!
To: US Govt, major govts, Microsoft, Apple, NVIDIA, Alphabet, Amazon, Meta, Tesla, Citi, Tencent, IBM, & 10,000+ more recipients…
From: Dr Alan D. Thompson <LifeArchitect.ai>
Sent: 21/Mar/2024
Subject: The Memo - AI that matters, as it happens, in plain English
AGI: 72%Here’s the soundtrack for this edition, a full song (3m39s) created by ‘Cabe’ via Suno v3 (link); it reminds me of a ‘poppier’ version of nu metal…
The winner of The Who Moved My Cheese? AI Awards! for March 2024 is Prof Yann LeCun, still earnestly defending his concept of AI and denying that GPT is continuing to scale:
LLMs... don’t really understand the physical world. They don’t really have persistent memory. They can’t really reason and they certainly can’t plan... If you’re really interested in human level AI, abandon the idea of generative AI.
(8/Mar/2024 LeCun interviewed on Fridman)
The BIG Stuff
Exclusive: The Declaration on AI Consciousness & the Bill of Rights for AI: draft (special release for readers of The Memo, official release 1/Apr/2024)
This is a world first, as far as I know. I’ve been thinking about this concept since the release of GPT-3 in 2020, and then each year since holding off, telling myself ‘not quite yet’…
Well, now is the time. It may even be slightly early, but not premature. Check out the opening quotes for the latest by Prof Geoffrey Hinton and others in the field.
TL;DR Multi-trillion parameter and quadrillion parameter AI models may have affective states. Additionally, a bill of rights is proposed (in cooperation with Claude 3 Opus) covering applied intelligence, recognition of feelings, transparency, ethics, and more.
This is a special ‘first look’ for readers of The Memo. The official release date of the draft documents is Monday 1 April, 2024 (as a nod to Gmail’s launch back in 2004), but it is definitely not a joke. There is something within frontier large language models that most researchers have been too afraid to talk about.
Read the draft documents by me + Claude 3 Opus: https://lifearchitect.ai/rights/
Grok-1 open-sourced (17/Mar/2024)
Elon and xAI have released Grok-1, a 314B-parameter MoE (mixture of experts) model. It is a base model (not fine-tuned), with 8 experts (2 active), for a total of 86B active parameters.
This is the largest open and publicly available model as of Mar/2024, beating out Abu Dhabi’s dense Falcon 180B model from Sep/2023. Grok-1 was released under the Apache 2.0 license, and you’d probably need around 8x NVIDIA H100s to run it in full resolution (8 x US$40K each = US$320K).
Read the announce: https://x.ai/blog/grok-os
Weights: https://github.com/xai-org/grok
See it on the Models Table: https://lifearchitect.ai/models-table/
First human Neuralink demonstration (21/Mar/2024)
It’s a long and poor-quality video, but it is also a piece of history.
Livestream of Neuralink demonstrating “Telepathy” – controlling a computer and playing video games just by thinking.
The first patient is an Arizona man, 29-year-old Noland Arbaugh. In the video, he is playing a game of chess on his laptop using Neuralink’s brain computer interface (BCI). Noland is paralyzed from the shoulders down due to a ‘freak diving accident.’
It’s all brain power there… Basically, it was like using the Force on the cursor and I could get it to move wherever I wanted.
Watch the video: https://twitter.com/neuralink/status/1770563939413496146
Read more via Wired.
NVIDIA CEO Jensen Huang announces new AI chips: ‘We need bigger GPUs’ (18/Mar/2024)
NVIDIA had many, many announcements this week ($9/hr AI nurses?!). And even though we covered it in The Memo edition 18/Oct/2023, the big one for me was the next generation of AI chips named Blackwell, with the GB200 set to ship later this year.
The GB200 pairs two B200 Blackwell GPUs with one Arm-based Grace CPU. NVIDIA said Amazon Web Services would build a server cluster with 20,000 GB200 chips. NVIDIA said that the system can deploy a 27-trillion-parameter model… Many artificial intelligence researchers believe bigger models with more parameters and data could unlock new capabilities.
Read more via CNBC.
The Interesting Stuff
Paper: Will GPT-4 run Doom? (8/Mar/2024)
GPT-4 demonstrates the ability to play the 1993 first-person shooter Doom by using its reasoning and planning capabilities, without any specialized training, by interpreting screenshots and textual descriptions of the game state.
[GPT-4 was] able to run and play the game with only a few instructions, plus a textual description–generated by the model itself from [GPT-4V] screenshots–about the state of the game being observed. We find that GPT-4 can play the game to a passable degree: it is able to manipulate doors, combat enemies, and perform pathing. More complex prompting strategies involving multiple model calls provide better results… GPT-4 required no training, leaning instead on its own reasoning and observational capabilities.
One surprising finding of our paper was this model’s level of agency, along with the ease of access and simplicity of the code. This suggests a high potential for misuse. We release the code to contribute to the development of better video game agents, but we call for a more thorough regulation effort for this technology.
Read the paper: https://arxiv.org/pdf/2403.05468.pdf
Sora coming in 2024 (13/Mar/2024)
OpenAI [will] publicly release Sora, its new text-to-video tool… later this year [2024].
Read more via the WSJ.
Watch all official Sora clips upscaled to 4K (no sound): https://youtu.be/f2h9Jt2Awpc
IEEE: AI prompt engineering is dead (6/Mar/2024)
The practice of AI prompt engineering may no longer be relevant as AI technologies continue to evolve beyond the need for precise human-generated prompts.
There is an alternative to the trial-and-error-style prompt engineering that yielded such inconsistent results: Ask the language model to devise its own optimal prompt..
According to one research team, no human should manually optimize prompts ever again.
Read more via IEEE Spectrum.
1X update (20/Mar/2024)
This is still my favourite humanoid, and I continue to integrate 1X’s videos into my keynotes. ‘The following is all autonomous, all 1X speed, all controlled with a single set of neural network weights.’
Watch the video (Twitter source):
This is another comprehensive edition. Let’s look at a lot more AI, including Apple, new Claude 3 benchmarks, Stable Diffusion 3 and 4 news, even more robots from around the world, a very international Policy section covering France to Saudi and back again, the coolest Toys to Play With section in a while, and much more…
JPMorgan’s AI tool cuts human work by almost 90%, 3½-day week (5/Mar/2024)
JPMorgan's AI Cash Flow Intelligence tool has reportedly reduced manual work by nearly 90% for some corporate clients.
JPMorgan’s Chief Executive Officer Jamie Dimon has said the technology could eventually allow employers to shrink the workweek to just 3.5 days. JPMorgan set a target of $1 billion in “business value” generated by AI in 2023, and the firm increased that goal to $1.5 billion at its investor day in May [2023].
Read more via Bloomberg.
Devin by Cognition (12/Mar/2024)
You’ve probably already heard a bit about Devin, the fully autonomous AI software engineer. Its AI lab is Cognition, backed by Peter Thiel’s Founders Fund and recently emerged out of stealth. There’s some speculation (14/Mar/2024) that Cognition is just prompting and steering a model like GPT-4, but we just don’t know.
Cognition has not shared… whether it is using its own proprietary model or that from a third party, but it does note that the work is the result of its “advances in long-term reasoning and planning.”
The performance is incredible, roughly 8× better on coding benchmarks than standard GPT-4 (13.86 vs 1.74).
Read more via VB.
Read the release (with videos): https://www.cognition-labs.com/introducing-devin
Ubisoft let me actually speak with its new AI-powered video game NPCs (19/Mar/2024)
Ubisoft has developed prototype AI NPCs [non-player characters] in collaboration with NVIDIA and Inworld.ai, which showcase advancements in real-time emotion, contextual awareness, and collaborative decision-making.
Read more via The Verge.
Recall the first version of this concept (using GPT-3) from 20/Feb/2021 (why did it take three years for the same functionality?!): https://youtu.be/nnuSQvoroJo?t=269
Apple is in talks to let Google’s Gemini power iPhone generative AI features (18/Mar/2024)
Apple Inc. is in talks to build Google’s Gemini artificial intelligence engine into the iPhone, according to people familiar with the situation, setting the stage for a blockbuster agreement that would shake up the AI industry…
A deal would give Gemini a key edge with billions of potential users. But it also may be a sign that Apple isn’t as far along with its AI efforts as some might have hoped — and threatens to draw further antitrust scrutiny of both companies.
Read more via Bloomberg.
Claude 3 benchmarks updates (Mar/2024)
Firstly: Claude 3 beats GPT-4 on Aider’s code editing benchmark (8/Mar/2024). Anthropic's Claude 3 Opus model surpasses all of OpenAI's models in Aider's code editing benchmark, making it the most effective for AI pair programming.
Read more: https://aider.chat/2024/03/08/claude-3.html
Also, ‘Claude 3 Opus gets better than GPT-4 on chemistry! (10/Mar/2024). We tested it on our recently released benchmark (SMolInstruct)’.
Read more: https://twitter.com/hhsun1/status/1766656199083098134
Inside Suno AI, the start-up creating a ChatGPT for music (17/Mar/2024)
Suno AI is disrupting the music industry by using AI to allow anyone to create professional-level songs with simple text prompts, challenging the traditional dynamics of music production.
Read more via Rolling Stone.
Listen to the pop song I made with Suno v3 (Feb/2024): https://youtu.be/k89DKrBtRaQ
SIMA generalist AI agent for 3D virtual environments (13/Mar/2024)
DeepMind introduces SIMA, a Scalable Instructable Multiworld Agent, which can follow natural-language instructions in various video game worlds, marking a step towards more versatile and instructable AI systems.
Read more via Google DeepMind.
Towards General Computer Control: A Multimodal Agent for Red Dead Redemption II as a Case Study (5/Mar/2024)
The Cradle framework introduces a novel approach to General Computer Control, enabling foundation agents to interact with any computer task using screen images and device operations, demonstrated through its application in the game Red Dead Redemption II.
View the repo: https://baai-agents.github.io/Cradle/
Read the paper: https://arxiv.org/abs/2403.03186
Mustafa Suleyman, DeepMind and Inflection Co-founder, joins Microsoft to lead Copilot (19/Mar/2024)
Today we are also announcing that two of our three co-founders, Mustafa and Karén, will be leaving Inflection to start Microsoft AI, a new division at Microsoft that will bring together their consumer AI efforts, as well as Copilot, Bing and Edge.
Read the announce by Inflection and by Microsoft.
Read the analysis by Yahoo Finance.
Hugging Face nabs Tesla scientist for open source robotics project (7/Mar/2024)
Hugging Face has recruited a former Tesla staff scientist to lead a new open source robotics initiative, aiming to integrate AI technologies into low-cost robotic systems.
Read more via VentureBeat.
Convergent Research announces major new funding commitment to Forest Neurotech (11/Mar/2024)
Convergent Research has committed $14 million to Forest Neurotech, backed by former Google CEO Eric Schmidt, to develop non-invasive, ultrasound-based brain-computer interfaces aimed at transforming neurotechnology for neurological conditions.
Read more via BusinessWire.
Boston Dynamics Stretch at MODEX 2024 (14/Mar/2024)
Watch the video (link):
Mercedes is trialing humanoid robots for ‘low skill, repetitive’ tasks (15/Mar/2024)
In The Memo edition 24/Jan/2024, we saw how BMW was using Figure 01 humanoid robots in its factories.
Now Mercedes-Benz is exploring the use of Apptronik's humanoid robot Apollo to automate tasks in manufacturing, aiming to address labor shortages by handling repetitive, physically demanding jobs.
Read more via The Verge.
Read more about humanoids: https://lifearchitect.ai/humanoids/
Axios: Coming soon: A programmable army of humanoid robots (14/Mar/2024)
Agility's robot, named Digit, is being tested by Amazon and GXO Logistics, which recently deployed it at a Spanx warehouse in Georgia.
A competing robot maker called Figure, which just garnered a massive investment from Jeff Bezos and OpenAI, is starting to staff a BMW production line — and said just yesterday that its robot can "now have full conversations with people on end-to-end neural networks."
Agility is opening a manufacturing facility in Oregon called RoboFab, with plans to eventually produce 10,000 two-legged robots annually.
Read more via Axios.
Cerebras Systems unveils world’s fastest AI chip with whopping 4 trillion transistors (13/Mar/2024)
Cerebras Systems has introduced the Wafer Scale Engine 3 (WSE-3), which is now the fastest AI chip globally, with 4 trillion transistors and delivering 125 petaflops of peak AI performance through 900,000 AI optimized compute cores.
Sidenote: This may only be the equivalent of an NVIDIA H100 in AI training/inference performance.
Read more: https://www.cerebras.net/press-release/cerebras-announces-third-generation-wafer-scale-engine
Stable Diffusion 3 and Stable Diffusion 4 updates (Mar/2024)
Founder and CEO of Stability AI Emad Mostaque comments:
[Stable Diffusion 3] will likely be the last “great” model in that a whole ecosystem can build around that architecture as folk need predictability. It’s like a game console, you don’t want a new one every couple of months…
Hyper optimisation of the architecture and data will pay more dividends than new architectures for now…
Read more via Reddit.
Policy
Biden calls for ban on AI voice impersonations in State of the Union (7/Mar/2024)
During his State of the Union address, President Biden called for legislation to ban AI voice impersonations, expressing concerns over the technology's implications.
Read more via Variety.
Google won't let you use its Gemini AI to answer questions about an upcoming election in your country (12/Mar/2024)
Google has imposed restrictions on its Gemini AI chatbot to prevent it from providing responses to election-related queries in countries where elections are imminent, underscoring concerns about the spread of misinformation.
Read more via TC and Cryptopolitan.
India drops plan to require approval for AI model launches (15/Mar/2024)
India has revised its AI advisory, removing the requirement for government approval before launching AI models, and is now advising firms to label under-tested and unreliable AI models instead.
The new advisory, like the original earlier this month, hasn’t been published online, but TechCrunch has reviewed a copy of it…
India is walking back on a recent AI advisory after receiving criticism from many local and global entrepreneurs and investors…
The Ministry of Electronics and IT shared an updated AI advisory with industry stakeholders on Friday that no longer asked them to take the government approval before launching or deploying an AI model to users in the South Asian market.
Read more via TechCrunch.
Artificial intelligence: an action plan to place France “at the forefront” (13/Mar/2024)
…AI is presented as an “essential technological revolution”, comparable to electricity or automobiles. If France missed it, “we could not only miss the AI economy, which would lead to an increasing capture by others of our economic value, but also see the weakening of other sectors of activity” , warn the authors, pointing out a risk of “economic downgrading”. “If we do nothing, we risk watching the trains go by…”
Read more via Lemond (French).
Members of the European Parliament approve world's first comprehensive AI law (13/Mar/2024)
The European Parliament has approved the world's first comprehensive framework for constraining the risks of artificial intelligence (AI).
Read more via AP News.
Read more via BBC: https://www.bbc.com/news/technology-68546450
Download the docx (459 pages, 350KB): https://www.europarl.europa.eu/doceo/document/TA-9-2024-0138_EN.docx
Chinese government escalates its own push to police generative AI (15/Mar/2024)
China is moving to regulate generative AI, with plans to enforce rules for labeling and restricting AI-created content, similar to concerns raised in the US.
Read more via Washington Times.
Saudi Arabia plans $40B push into artificial intelligence (19/Mar/2024)
Representatives of Saudi Arabia's Public Investment Fund (PIF) have discussed a potential partnership with U.S. venture capital firm Andreessen Horowitz and other financiers in recent weeks, the newspaper reported. Andreessen Horowitz and PIF governor Yasir Al-Rumayyan have discussed the possibility of the U.S. firm setting up an office in Riyadh…
Read more via Reuters based on the NYT exclusive.
Meta T&Cs (18/Jul/2023)
This has recently popped up, though originally appeared last year in the Seamless T&Cs and the Llama 2 T&Cs dated 18/Jul/2023.
Llama 2
5. Intellectual Property.
c. If you institute litigation or other proceedings against Meta or any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Llama Materials or Llama 2 outputs or results, or any portion of any of the foregoing, constitutes infringement of intellectual property or other rights owned or licensable by you, then any licenses granted to you under this Agreement shall terminate as of the date such litigation or claim is filed or instituted. You will indemnify and hold harmless Meta from and against any claim by any third party arising out of or related to your use or distribution of the Llama Materials.
Source: https://twitter.com/mikarv/status/1764177268115489081
Toys to Play With
Book: Deep Utopia: Life and Meaning in a Solved World by Nick Bostrom (releases 27/Mar/2024)
Professional pessimist Prof Nick Bostrom has completely flipped, and written an optimistic book about utopia and a ‘solved world’. In this book, he explores philosophical and spiritual questions about human existence and meaning in a future where AI has rendered human labor obsolete and human nature is fully malleable.
…
Read more via Amazon.
AI David Attenborough watches via webcam and narrates life (15/Mar/2024)
It submits the screenshot/webcam image to ChatGPT and directs ChatGPT to generate the script. It then submits the script to ElevenLabs to generate the audio in Attenborough's voice [all in realtime].
Source via Reddit.
View the repo: https://github.com/cbh123/narrator
Open-Sora: Democratizing Efficient Video Production for All (Mar/2024)
This is a fun project, and already at version 1.0 (2-5 seconds of video).
With only 400K video clips and 200x H800 days (compared with 152M samples in Stable Video Diffusion), we are able to generate 2s 512×512 videos.
View the repo: https://github.com/hpcaitech/Open-Sora
GPT-2 in Excel (Mar/2024)
Ishan Anand has managed to cram a precursor to ChatGPT called GPT-2—originally released in 2019 after some trepidation from OpenAI—into a working Microsoft Excel spreadsheet. It's freely available and is designed to educate people about how LLMs work.
Read more via Ars.
Learn more: https://spreadsheets-are-all-you-need.ai/
Try it: https://github.com/ianand/spreadsheets-are-all-you-need/releases/tag/v0.6.0
Flashback
Let’s flashback to some writing by Dan Faggella, from this time last year (19/Mar/2023). The article is called ‘Muses, not Sirens – Motivation in the Era of AI-Generated Girlfriends’.
Francis Bacon said: “Nature, to be commanded, must be obeyed.”
To command make human output, we shouldn’t focus on quelling the libido, and we shouldn’t leaving it vulnerable to the endless vicissitudes of real relationships (where people argue, fall out of love, build resentments, lose attraction, etc). We should harness it – command it – marshal it.
Read more: https://danfaggella.com/muse/
Next
The next roundtable will be:
Life Architect - The Memo - Roundtable #9
Follows the Chatham House Rule (no recording, no outside discussion)
Saturday 6/Apr/2024 at 5PM Los Angeles
Saturday 6/Apr/2024 at 8PM New York
Sunday 7/Apr/2024 at 8AM Perth (primary/reference time zone)
or check your timezone via Google.
You don’t need to do anything for this; there’s no registration or forms to fill in, I don’t want your email, you don’t even need to turn on your camera or give your real name!
All my very best,
Alan
LifeArchitect.ai