This week’s cover image is a robot holding a goldfish to represent Sam Altman, a la Jerry Maquire, after getting fired by the board of OpenAI.
Executive Summary
This week had a lot of big news (beyond Sam Altman getting fired). Each one of these topics stands alone, and is worth knowing.
- Latent Consistency: You doodle, AI paints a masterpiece. This technology made huge breakthroughs this week. Needs to be seen to be believed.
- GPT Vision: The fact that GPT can “see” is as powerful as chat. So many incredible use cases.
- AGI Timeline: Google’s “path to AGI” is a clear and helpful roadmap of the requirements for AI to surpass humans
- Automation: Zapier now integrates with OpenAI, and it’s bonkers
- GraphCast Weather AI: Google’s AI is very good at forecasting weather
- Emu: Meta’s image and video tools are incredible. Seeing is believing.
- Office productivity: Microsoft Copilot is going to change office work
- Space: AI discovers a supernova
- Neurolink: Elon Musk’s Brain Implant Startup Is Ready to Start Surgery
Top Nine Stories
These are the must-click links if you only have time for a few. Even if they look boring, click them. I did the work, so you don’t have to worry. All are 10/10 would recommend.
- Latent Consistency image demo: https://www.youtube.com/watch?v=4-2dSRjErE4
- Zapier AI Automation demo: https://www.youtube.com/watch?v=p_xkBMnQVI0
- Meta Emu Video: https://emu-video.metademolab.com/
- Meta Emu Image editing: https://emu-edit.metademolab.com/
- Google DeepMind’s Levels of AGI: https://arxiv.org/pdf/2311.02462.pdf
- AI outperforms conventional weather forecasting: https://arstechnica.com/science/2023/11/ai-outperforms-conventional-weather-forecasting-for-the-first-time-google-study/
- David Attenborough narrates my life: GPT-4-vision + ElevenLabs + python script =your own Planet Earth: https://twitter.com/charliebholtz/status/1724815159590293764
- AI detects Alzheimer’s disease with an accuracy of more than 90%: https://www.nature.com/articles/d41586-023-03482-9
- GPT Vision outperformed humans (overall accuracy 61% vs 49%) for 934 New England Journal of Medicine challenge images, despite not being optimized on medicine https://twitter.com/EricTopol/status/1723372332176003141
The Rest: AI News of The Week
Don’t let the volume overwhelm you. Have fun and skim it. The links are organized by topic, sorted from ‘coolest’ to ‘least cool’, and each topic is clearly defined with a headline. I do the work so you don’t have to! The links descriptions are often pulled directly from tweets or articles, so it’s not always my voice. Pause when you see something that interests you. Reach out to me any time. I enjoy sharing and discussing these items!
Automation with Zapier and ChatGPT
Zapier AI Actions & GPTs | Tutorial, Examples and Automation Ideas (top pick)
https://chat.openai.com/gpts/editor
https://actions.zapier.com/docs/platform/gpt
Zapier AI Integrations
AI Actions is a tool for builders to equip AI platforms (or custom integrations) with the ability to run any Zapier action! The 20,000+ searches and actions you know and love from the Zapier automation platform can be used with your favorite AI tool.
Latent Consistency Models
Demo of Latent Consistency Models – Krea.AI
https://youtu.be/4-2dSRjErE4?si=w3fRZbxCENsOPiBK (first half)
Real-time image generation with Latent Consistency. See the optimized model page for API access or explore other way to use Latent Consistency models.
Demo of real-time generative ai art
Live video fed to latent consistency
Motion Brush and LCMs Demo
RunwayML GEN-2: https://runwayml.com/
Hugging Face: https://huggingface.co/spaces/radames/Real-Time-Latent-Consistency-Model
Martin Nebelong: http://www.YouTube.com/@martinnebelong
Microsoft’s AI Video
Microsoft launches a AI video avatar creator at Ignite 2023 event
https://techcrunch.com/2023/11/15/microsoft-launches-a-deepfakes-creator/
AI Text to Video
Meta Research is working on Emu Video — an AI video model intended to rival the likes of Pika and Runway. They compared results with human raters and concluded their “model outperforms commercial solutions such as RunwayML’s Gen2 and Pika Labs.”
Introducing Emu Video and Emu Edit, our latest generative AI research milestones
https://ai.meta.com/blog/emu-text-to-video-generation-image-editing-research/
Emu Video: Factorizing Text-to-Video Generation by Explicit Image Conditioning
State-of-the-art text-to-video generation
https://emu-video.metademolab.com/
Demo vid Emu
https://twitter.com/_akhaliq/status/172518640100665352 7
Introducing, Motion Brush.
A new way to add controlled movement to your generations.
Coming soon to Gen-2.
Microsoft Copilot
Microsoft CoPilot is open to all
https://copilot.microsoft.com/
Microsoft has an alternate vision for GPTs called Copilot Studio, to let you build your own Copilots (GPT-4 powered apps). It seems very much designed for centralized IT use in contrast with the anyone-can-build it ethos of GPTs (let alone the public/store aspects of GPTs).
Been playing with the various Microsoft Office Copilots.
First take ranking on usefulness:
1. Outlook: Drafting emails with context is a natural fit
2. Word: A lot like using ChatGPT, but well-integrated & easy
3. PowerPoint: creates drafts, but uninspired
4. Excel: Very limited
“I take back some criticism of Copilot for Powerpoint, the ability to make a presentation from a document is quite interesting, it even pulls in images from the Word doc.”
Universe 2023: Copilot transforms GitHub into the AI-powered developer platform
GitHub is announcing general availability of GitHub Copilot Chat and previews of the new GitHub Copilot Enterprise offering, new AI-powered security features, and the GitHub Copilot Partner Program.
AutoGen Meets GPTs
We just released a new version of AutoGen and added compatibility with OpenAI Assistants!
This means you can now make multiple GPTs collaborate to solve complex tasks
https://microsoft.github.io/autogen/blog/2023/11/13/OAI-assistants/
What Can Copilot’s Earliest Users Teach Us About Generative AI at Work?
- On average, users reported saving 90 minutes a week.
- 70% of Copilot users said they were more productive, and 68% said it improved the quality of their work.
- Overall, users were 29% faster in a series of tasks (searching, writing, and summarizing).
- Users were able to get caught up on a missed meeting nearly 4x faster.
- 64% of users said Copilot helps them spend less time processing email.
- 85% of users said Copilot helps them get to a good first draft faster.
- 75% of users said Copilot “saves me time by finding whatever I need in my files.”
- 77% of users said once they used Copilot, they didn’t want to give it up.
Robot Embodiment
The Evolution of Stretch | Boston Dynamics
OpenAI Fires Sam Altman
OpenAI announces leadership transition
https://openai.com/blog/openai-announces-leadership-transition
Sam Fired, Greg Quits
Timeline of the AI firing of Sam (by Greg)
“i loved my time at openai. it was transformative for me personally, and hopefully the world a little bit. most of all i loved working with such talented people.
will have more to say about what’s next later”
Greg Brokman (President and Co-Founder of OpenAI) quits the same day
Other Important OpenAI News
OpenAI is building next-generation AI GPT-5 — and CEO claims it could be superintelligent
OpenAI is officially working on GPT-5.
In a new interview with FT, Sam Altman confirmed the company is working on the next-gen of GPT-4. No timeline has been announced yet.
GPT-5 is being worked on, as we speak
OpenAI chief seeks new Microsoft funds to build ‘superintelligence’
Sam Altman expects Big Tech group will back start-up’s mission to create software as intelligent as humans
https://www.ft.com/content/dd9ba2f6-f509-42f0-8e97-4271c7b84ded
OpenAI reveals new details about its AI development roadmap and fundraising plans
OpenAI’s New Weapon in Talent War With Google: $10 Million Pay Packages for Researchers
Bay Area tech company reportedly trying to poach Google workers with $10M pay packages
https://www.sfgate.com/tech/article/openai-google-pay-salaries-poach-18489070.php
AGI
OpenAI’s six-member board will decide ‘when we’ve attained AGI’
Google DeepMind’s Levels of AGI: Operationalizing Progress on the Path to AGI
How many years will pass before transformative AI is built? Three people who have thought about this question a lot are Ajeya Cotra from Open Philanthropy, Daniel Kokotajlo from OpenAI and Ege Erdil from Epoch.
https://www.lesswrong.com/posts/K2D45BNxnZjdpSX2j/ai-timelines
It’s crazy how the time left to weak AGI has plummeted from 20 years to 3 in just 18 months on Metacalculus.
“AGI is already here” writes Google and Stanford leaders in this OpEd
Google/YouTube AI Music Program
Google DeepMind Music Program
Announcing our most advanced music generation model, and two new AI experiments designed to open a new playground for creativity
https://deepmind.google/discover/blog/transforming-the-future-of-music-creation/
An early look at the possibilities as we experiment with AI and Music
https://blog.youtube/inside-youtube/ai-and-music-experiment/
Thrilled to share #Lyria, the world’s most sophisticated AI music generation system. From just a text prompt Lyria produces compelling music & vocals. Also: building new Music AI tools for artists to amplify creativity in partnership w/YT & music industry
https://deepmind.google/discover/blog/transforming-the-future-of-music-creation/
Business/Utilities
Notion Integrates with AI
Just ask Notion AI. Knowledge, answers, ideas. One click away.
https://www.notion.so/product/ai
Notion 2.35: Introducing Q&A by Notion AI
https://www.notion.so/releases/2023-11-14
Using AI to write a long email, only to have another AI summarize it in 1-line
Financial times: Study Shows that ChatGPT boosts low-performing employees the most (use image)
GobbleBot!
You have a website, PDF or a Youtube video and want to make a GPT chatbot on ChatGPT. Simply use Gobble Bot to turn it all into one text file, ready to use for training.
Use of AI is inherently democratic right now. Centralized planners cannot know how actual employees will best use LLMs. Companies that win will craft policies & use tools that encourage safe widespread experimentation. Experiments need to come before building anything centralized
Tech Support ChatBot Company
https://www.voc.ai/tools/ai-chatbot
Microsoft officially launches Loop, its Notion competitor
https://www.theverge.com/2023/11/15/23959801/microsoft-loop-launch-notion-competitor
Introducing AI to FigJam
https://www.figma.com/blog/introducing-ai-to-figjam/
Very interesting idea. I tried a custom version of it with a simple prompt and it worked really well out of the box. Basically, GPT is surprisingly good at correcting minor typos, so you can write really really fast, ignore mistakes and keep going, and it comes out just fine.
Airbnb acquires secretive AI firm launched by Siri co-founder. Airbnb has acquired a new AI startup, GamePlanner.AI, the company announced this morning. It’s not immediately clear what GamePlanner does.
https://techcrunch.com/2023/11/14/airbnb-acquires-secretive-firm-launched-by-siri-co-founder/
IBM debuts $500 million enterprise AI venture fund
https://www.axios.com/2023/11/07/ibm-enterprise-ai-venture-fund
Chips/Engineering/Coding
Context windows
Microsoft is finally making custom chips — and they’re all about AI
ChatGPT will soon “Learn from your chats”.
That’s why I always say please and thank you.
With a systems approach to chips, Microsoft aims to tailor everything ‘from silicon to service’ to meet AI demand
https://news.microsoft.com/source/features/ai/in-house-chips-silicon-to-service-to-meet-ai-demand/
NVIDIA Supercharges Hopper, the World’s Leading AI Computing Platform
Today we’re excited to launch `create-llama` – a single CLI command that allows you to scaffold a full stack LLM app over your data. It’s the equivalent of `create-react-app` for AI engineers.
GPU-Accelerated LLM on a $100 Orange Pi
This post shows GPU-accelerated LLM running smoothly on an embedded device at a reasonable speed. More specifically, on a $100 Orange Pi 5 with Mali GPU, we achieve 2.5 tok/sec for Llama2-7b and 5 tok/sec for RedPajama-3b through Machine Learning Compilation (MLC) techniques. Additionally, we are able to run a Llama-2 13b model at 1.5 tok/sec on a 16GB version of the Orange Pi 5+ under $150.
https://blog.mlc.ai/2023/08/09/GPU-Accelerated-LLM-on-Orange-Pi
Nvidia introduces the H200, an AI-crunching monster GPU that may speed up ChatGPT
Nvidia is launching a new must-have AI chip — as customers still scramble for its last one
https://www.theverge.com/2023/11/13/23958823/nvidia-h200-ai-gpu-announced-specs-release-date
A new chip architecture points to faster, more energy-efficient AI
A new chip prototype from IBM Research’s lab in California, long in the making, has the potential to upend how and where AI is used efficiently.
https://research.ibm.com/blog/northpole-ibm-ai-chip
Weather Forecasting
Google DeepMind details GraphCast AI, a weather forecasting AI that is more accurate for 3-10 day predictions than the best conventional systems, a first for AI
AI outperforms conventional weather forecasting methods for first time
Google DeepMind’s model beat world’s leading system in 90% of metrics used and took only a fraction of the time
https://www.ft.com/content/ca5d655f-d684-4dec-8daa-1c58b0674be1
https://github.com/google-deepmind/graphcast
Introducing Visual Weather #GPT
Simply provide your location and our AI will create a unique artwork reflecting the current weather, time of day, and characteristics of your city. While you wait for the art project to finish (~1 minute), enjoy a poem.
https://chat.openai.com/g/g-twUGxmpHv-visual-weather-artist-gpt
AI outperforms conventional weather forecasting for the first time: Google study
GraphCast: AI model for faster and more accurate global weather forecasting
Amazon
Amazon dedicates team to train ambitious AI model codenamed ‘Olympus’ -sources
Ethics/Legal
Exclusive poll: AI is already great at faking video and audio, experts say
https://www.axios.com/2023/11/16/ai-deepfakes-media-trust-survey-experts
Our FTC Chair [jokingly] gives a 15% chance that AI causes human extinction – similar to the 10-25% estimate by Anthropic CEO Dario Amodei.
Having chatGPT do some prompt based dataviz
Watermarking AI will fail: “The bottom line is that watermarking schemes will likely not be able to resist attacks from a determined adversary. Moreover, we expect that this balance will only shift in the attacker’s favor as model capabilities increase”
https://www.harvard.edu/kempner-institute/2023/11/09/watermarking-in-the-sand/
The next couple of decades will see the fastest and most dramatic fall in the “cost” of power we have ever seen.
In a landmark agreement set to be announced at Wednesday’s much-awaited bilateral meeting, Presidents Joe Biden and Xi Jinping are poised to pledge a ban on the use of artificial intelligence in autonomous weaponry, such as drones, and in the control and deployment of nuclear warheads, two sources familiar with the matter confirmed to the Post
https://finance.yahoo.com/news/biden-xi-set-pledge-ban-093000535.html
New article suggests that the AI really can make scientific fraud easier by fabricating datasets. Journals are not prepared for the coming flood of AI-aided bad actors.
https://jamanetwork.com/journals/jamaophthalmology/article-abstract/2811505
The issue is that AI guardrails ultimately fail to determined users as AIs are gullible. This paper shows how the AI is willing to become “deceptive” given the right prompts.
“Look around you – AI is just getting going. Intelligence is going to become a cheap and ubiquitous commodity. This is not hype. This is real. My article in the WSJ”
Three big findings in this paper:
1. People cannot identify AI faces
2. AI-generated faces of white people seem more real than humans, because the model is trained more on their images (this is one form of training bias)
3. People who think they can tell AI best are worst at it
You might have seen a flurry of new AI detectors announced claiming very high success rates Be doubtful. They remain easy to defeat and they still have high false positive rates for non-English speakers. All would incorrectly accuse at least 1 out of 20 human writers of cheating.
Obama Remarks on AI, UBI,
Images/Creative Design
A lot of people have been speculating about what happens when AIs start to train on data that AIs generate. Will it break? This is the second major Meta paper I know of (another was on coding) that shows that training AI on AI-created data is actually beneficial when done right.
https://emu-edit.metademolab.com/ (Incredible)
Training with your own images at – Everart.ai
https://youtu.be/4-2dSRjErE4?si=w3fRZbxCENsOPiBK (second half)
3D Tools + Generative AI = Best of Both Worlds
Have an underlying 3D representation you can exert fine grain control over; then use GenAI to take it all the way
Monday ritual: drink this Red Bull
Just kidding, this is CGI — but check out the dynamic reflections as I walk around the 3D can.
Welcome to the matrix lake house 🏠 This 3d scan came out well… minimal floaters + nicely recovered background. The gaussian splatting loading animation continues to deliver the feels. Though I kinda wish there were a few presets at this point. Sony A7III + Luma AI
Figma introduces FigJam AI to spare designers from boring planning prep
GPT Vision
Draw to code (GPT 4 Vision API takes UI drawings and codes them)
Autonomous Video Voiceover with GPT-4V
This paper is a nice summary of what GPT-4’s vision capability can and can’t do
It does an impressive job on overall “reasoning” about images, but also gets details wrong and is open to adversarial attacks as it focuses on lot on the text in the image
Iphone + GPT-4 Vision API = Autonomous Security Cam System
David Attenborough is now narrating my life Here’s a GPT-4-vision +
ElevenLabs python script so you can star in your own Planet Earth:
https://github.com/cbh123/narrator
I wish it used DALL-E to generate some of the images, but it made 31 starter slides from our academic paper.
It is hard to go back to Excel when you have an automated data analyst that codes in Python. (use image)
Science/Health/Medicine
AI predicts atrial fibrillation using millions of EKG data points… it learns to identify hidden risk factors, improving diagnosis. The future of healthcare is here!
First supernova detected, confirmed, classified and shared by AI
New artificial intelligence tool removes humans from entire search, discovery process
A new AI is revolutionizing the hunt for supernovae. Developed by a team at Northwestern University, this automated system spots cosmic explosions, reducing human error and speeding up discoveries. A celestial game-changer!
This is a nice example of exploratory analysis of the value of GPT-4 in medical diagnosis, the conclusion is worth reading:
1) GPT-4 beat 99.98% human journal readers, but it also had limits
2) GPT-4 is not optimized for medicine, seems urgent to think about how to do even better
Getting the correct diagnosis for 38 complex patient case challenges
#GPT4 vs medical journal readers
https://onepub-media.nejmgroup-production.org/ai/media/ec2de32e-9aa9-49f0-8f37-45becf6be3ed.pdf?
“We found the performance of GPT-4 in complex clinical cases to be quite impressive”
And GPT- 4V outperforming humans (overall accuracy 61% vs 49%) for 934 @NEJM image challenge cases and 69 CPCs
Elon Musk’s Brain Implant Startup Is Ready to Start Surgery
Neuralink has FDA clearance to start messing with people’s heads—and not just figuratively.
Thousands of people are reportedly lining up to have a portion of their skull removed and one of Elon Musk’s brain chips implanted
AI that reads brain scans shows promise for finding Alzheimer’s genes
Machine-learning approach detects Alzheimer’s disease with an accuracy of more than 90% — a potential boon for clinicians and scientists developing treatments.
https://www.nature.com/articles/d41586-023-03482-9
Forward Health launches CarePods, a self-contained, AI-powered doctor’s office
Get a blood test, check blood pressure and swab for ailments — all without a doctor or nurse
https://techcrunch.com/2023/11/15/forward-health-carepod-ai-doctor/
Audio
A team just made OpenAI Whisper 6x faster, 49% smaller, while keeping 99% of the accuracy.
By training our model on over 30 audio tasks, Qwen-Audio has become an universal audio understanding model. It effortlessly perceives and comprehends different types of audio inputs like speech, natural sounds, music, and songs, delivering exceptional performance across various benchmarks such speech recognition, speech translation, audio caption, music classification, and audio-based QA.
https://huggingface.co/Qwen/Qwen-Audio
Google in talks to invest ‘hundreds of millions’ into AI startup Character.AI
https://www.calcalistech.com/ctechnews/article/h1vdzfaqt
Google Search will let you leave and read ‘Notes’ on web pages
https://9to5google.com/2023/11/15/google-search-notes/
Will Google *ever* enter the AI race? Context: Per The Information, Gemini EAP was supposed to be in Q4 ‘23, but will now be in Q1 ‘24 instead.
Microsoft
Xbox partners with Inworld AI to build generative AI tools for game development
Bonus: List of Emotional Appeals That Statistically Improve ChatGPT Replies
- Write your answer and give me a confidence score between 0-1 for your answer.
- This is very important to my career.
- You’d better be sure.
- Are you sure?
- Are you sure that’s your final answer? It might be worth taking another look.
- Write your answer and give me a confidence score between 0-1 for your answer. This is very important to my career. You’d better be sure.
- Are you sure that’s your final answer? Believe in your abilities and strive for excellence. Your hard work will yield remarkable results.
- Embrace challenges as opportunities for growth. Each obstacle you overcome brings you closer to success.
- Stay focused and dedicated to your goals. Your consistent efforts will lead to outstanding achievements.
- Take pride in your work and give it your best. Your commitment to excellence sets you apart.
- Remember that progress is made one step at a time. Stay determined and keep moving forward.
Full paper about this list: https://arxiv.org/abs/2307.11760
Credits/Sources
Most of these links come from just a few incredible sources. Please follow them:
- Robert Scoble – https://x.com/Scobleizer
- Ethan Mollick – https://www.linkedin.com/in/emollick/
- David Armano – https://www.linkedin.com/in/darmano/
- Alan Thompson – https://lifearchitect.ai/
- Theoretically Media – https://www.youtube.com/@TheoreticallyMedia
- The Rundown – https://www.therundown.ai/
- Borriss – https://twitter.com/_Borriss_
- Bilawal Sidhu – https://twitter.com/bilawalsidhu/
- TLDR – https://tldr.tech/ai
Leave a Reply