Ethan B. Holland

Over 57,800 manually organized AI links and counting

AI News #9: Week Ending 12/01/2023 with Executive Summary and Top 8 Stories

For this week’s cover, I upscaled a low-res screengrab from the classic video game, GoldenEye 64, using Magnific AI and named it GoldenAI. Magnific is an incredible tool.

Executive Summary

Another week with a lot of big news, but this week’s theme is video, video, video.  Each one of these topics stands alone and is worth knowing.  See examples in the top stories section.

  • Video Avatars: HeyGen’s video clones are sneaking up on reality
  • Image and Video Upscaling: Magnific AI’s upscaler can transform AI-generated images and old movie frames into photorealistic masterpieces.
  • Text to Video:  A new company called Pika is taking text to video to the next level.  The old favorite, Runway AI, is one week into enabling pan, tilt, and zoom and examples are flowing in.  Stable Diffusion video launched last week, and initial examples are looking fantastic.  See links below.
  • Warfare: The US is on track to have fully autonomous lethal weapons in the future
  • Audio: Chinese company, Qwen, debuts an incredible multilingual speech recognition and translation tool that can discern gender, emotions, accent, age, context, and ambient sound such as music and surroundings.  See top stories.
  • Science: GPT4 is impressively strong at radiology. Google discovers 2.2 million crystal structures. 
  • Amazon AI: Amazon debuts a new LLM named “Q”, and is a bit of a dud
  • AI Image “More” Trend: An entertaining trend repeatedly asks AI to make an image “more” of something to see how the AI takes it to the extreme.

Top Stories

These are the links to click if you only pick 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.

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!

Images

Magnific AI Upscaler is incredible

Incredible Examples of Magnific Upscaling (click them all!)

Magnific.AI

https://magnific.ai/

Stability AI SDXL Turbo

Introducing SDXL Turbo: A Real-Time Text-to-Image Generation Model

https://stability.ai/news/stability-ai-sdxl-turbo

Try it here: https://clipdrop.co/stable-diffusion-turbo 

Report: Stability AI Positioning Itself for Acquisition

Report: Stability AI Positioning Itself for Acquisition

Krea AI is another upscaler

Fun trend… “More”… where you tell Dall-E to keep making an image “more” of something.

A cat, getting progressively successful in life 

Making the toddler run faster

Make a bedroom cozier

Spicy ramen, but then it gets spicier and spicier

Segmentation-Based Parametric Painting: Our method turns images into paintings of different styles. 

https://manuelladron.github.io/semantic_based_painting/

Coca-Cola sparks Diwali celebrations with personalised AI-generated wish cards

https://bestmediainfo.com/2023/11/cocacola-sparks-diwali-celebrations-with-personalised-aigenerated-wish-cards

Realtime generative AI art is here thanks to LCM-LoRA

Realtime generative AI art is here thanks to LCM-LoRA

https://huggingface.co/spaces/ilumine-AI/LCM-Painter

Video

HeyGen’s video avatars are getting incredibly realistic

This is one that’s not polished…an example of the “out of the box” functionality

AI video startup HeyGen can reportedly now generate avatars from smartphone clips in just 5 minutes.

Pika Just Dropped The Mic On AI Video

https://youtu.be/CHoOdPVSEsE?si=Ofn6cTkHSpx1LbDx

Introducing Pika 1.0, the idea-to-video platform that brings your creativity to life.

Create and edit your videos with AI.

https://pika.art

Generative Video Startup Pika Labs Launches Version 1.0, Raises $55 Million in Funding

https://petapixel.com/2023/11/28/generative-video-startup-pika-labs-launches-version-1-0-raises-55-million-in-funding/

Stunning Breakthroughs in AI Creativity (worth watching)

https://youtu.be/yfxZKoTOka0?si=xWJ0b6Chf8-bTF1C

The Future of AI Video Has Arrived! (Stable Diffusion Video Tutorial/Walkthrough)

https://youtu.be/82l0DsbLHhY?si=qS_b75wnwqtaVBFB

Meet the first Spanish AI model earning up to €10,000 per month

https://www.euronews.com/next/2023/12/02/meet-the-first-spanish-ai-model-earning-up-to-10000-per-month

Kandinsky Video is a text-to-video generation model, which is based on the FusionFrames architecture, consisting of two main stages – keyframe generation and interpolation. Our approach for temporal conditioning allows us to generate videos with high-quality appearance, smoothness and dynamics.

https://ai-forever.github.io/kandinsky-video

Runway Pan and Tilt Examples:

Runway Motion Brush Examples:

Health/Medicine/Science

“Surprisingly, we found radiology report summaries generated by GPT-4 to be comparable and, in some cases, even preferred over those written by experienced radiologists”

GPT-4’s potential in shaping the future of radiology

AI system self-organizes to develop features of brains of complex organisms

https://www.cam.ac.uk/research/news/ai-system-self-organises-to-develop-features-of-brains-of-complex-organisms

Google AI and robots join forces to build new materials

https://www.nature.com/articles/d41586-023-03745-5

Gen AI for the Genome: LLM Predicts Characteristics of COVID Variants

https://blogs.nvidia.com/blog/generative-ai-covid-genome-sequences/

Millions of new materials discovered with deep learning

https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/

AI under the microscope: the algorithms powering the search for cells

https://www.nature.com/articles/d41586-023-03722-y

Google’s DeepMind finds 2.2M crystal structures in materials science win

https://arstechnica.com/ai/2023/11/googles-deepmind-finds-2-2m-crystal-structures-in-materials-science-win/

OpenAI

OpenAI’s Mission – A reminder from Ethan Mollick, professor at Wharton

“From Altman on down, the people at OpenAI are remarkably straightforward about their goals: building semi-autonomous AI agents in the near term & then building AGI.  You don’t need to believe that they can do it, but you should probably plan for a world where they do.”

ChatGPT turns1

“a year ago tonight we were probably just sitting around the office putting the finishing touches on chatgpt before the next morning’s launch. what a year it’s been…”

Sam Altman returns as CEO, OpenAI has a new initial board

Official messages from Sam and Bret (board chair) are worth reading

https://openai.com/blog/sam-altman-returns-as-ceo-openai-has-a-new-initial-board

Sam Altman on being fired and rehired by OpenAI

https://www.theverge.com/2023/11/29/23982046/sam-altman-interview-openai-ceo-rehired

Helen Toner resigns from the OpenAI board

Inside OpenAI, a rift between billionaires and altruistic researchers unraveled over the future of artificial intelligence

https://www.abc.net.au/news/2023-11-26/openai-sam-altman-board-inside-the-chaotic-week/103149570

Perplexity AI

Perplexity AI unveils ‘online’ LLMs that could dethrone Google Search

Perplexity AI unveils ‘online’ LLMs that could dethrone Google Search

https://blog.perplexity.ai/faq/what-is-copilot

Introducing PPLX Online LLMs 

https://blog.perplexity.ai/blog/introducing-pplx-online-llms

Perplexity AI

https://www.perplexity.ai/

Agents

Send a photo of an item and ask it to find a less expensive alternative.

Upload a photo of a deal and check if it’s the best price for the item

Mastercard just rolled out a new AI tool called Shopping Muse.

The tool is meant to inspire and advise shoppers this holiday season.

Designed to mimic an in-store helper through natural language, Muse makes product matches based on profiles and past purchases.

VR/AR

Real time text to 3D generation (must watch)

Check out SDXL-Turbo + Stable Video AI in action

The Varjo XR4’s passthrough is a spiritual experience. 

On the Focal edition the cameras autofocus giving a surreal-realism. I’m blown away.

Gaussian Shell Maps are a new neural scene representation that connects fields and 3D Gaussians. This representation unlocks the full potential of 3D Gaussian splatting for generative AI applications, such as 3D avatar generation.

CGI Mixed Reality Shootout at the CyberTruck Corral

Apple’s Vision Pro will begin mass production in December, and despite reports Apple has been aggressively relocating its supply chain elsewhere, China suppliers will account for 60% of the device, media report, noting it costs US$1,700 per unit to make, with China’s LY iTech accounting for $250-$300 of that amount, and Luxshare Precision around $150-$200. The supply next year is seen at 400,000 units. https://twitter.com/dnystedt/status/1729690672582832429

Will the internet become VR? I’ve always been skeptical, but perhaps this is the next chapter…
https://www.newyorker.com/magazine/2023/12/04/how-jensen-huangs-nvidia-is-powering-the-ai-revolution 

Animatable Gaussians: Learning Pose-dependent Gaussian Maps

for High-fidelity Human Avatar Modeling

https://animatable-gaussians.github.io

AR Sports 

Super cool demo of watching AR sports on a table

A multimodal LLM that can discern poses, for remapping

LumaLabsAI has rolled out improved Gaussian Splats — now with less floaters and better reconstruction of distant background elements.

Audio

Powerful speech recognition (gender, emotions, accent, age, context, surroundings, content, music ID) and translation from Qwen

https://qwen-audio.github.io/Qwen-Audio/

Seamless: Multilingual Expressive and Streaming Speech Translation

https://ai.meta.com/research/publications/seamless-multilingual-expressive-and-streaming-speech-translation/

Ethics/Legal/Politics

World’s first AI minister likens risk of overregulation to calligraphers that kept the printing press out of the Middle East for nearly 200 years

https://archive.md/Vd6Yr#selection-705.0-705.145

The GPT to rule them all: Training for one trillion parameter model backed by Intel and US government has just begun

https://www.techradar.com/pro/the-gpt-to-rule-them-all-training-for-one-trillion-parameter-model-backed-by-intel-and-us-government-has-just-begun

Pentagon’s AI initiatives accelerate hard decisions on lethal autonomous weapons

https://apnews.com/article/us-military-ai-projects-0773b4937801e7a0573f44b57a9a5942

Microsoft: Our investment in AI infrastructure, skills and security to boost the UK’s AI potential 

https://blogs.microsoft.com/on-the-issues/2023/11/30/uk-ai-skilling-security-datacenters-investment/

ChatGPT’s training data can be exposed via a “divergence attack”

The researchers employed a methodology involving extensive token generation from different models, followed by a comparison of these tokens with the respective training datasets to identify instances of direct memorization.

ChatGPT’s training data can be exposed via a “divergence attack”

The Biden administration has forced a Saudi Aramco venture capital firm to sell its shares in a Silicon Valley AI chip startup backed by OpenAI co-founder Sam Altman

https://www.bloomberg.com/news/articles/2023-11-30/us-compels-saudi-fund-to-exit-ai-chip-startup-backed-by-altman?utm_campaign=socialflow-organic

Inside U.S. Efforts to Untangle an A.I. Giant’s Ties to China

US, Britain, other countries ink agreement to make AI ‘secure by design’

https://www.reuters.com/technology/us-britain-other-countries-ink-agreement-make-ai-secure-by-design-2023-11-27/

Sports Illustrated Published Articles by Fake, AI-Generated Writers

https://futurism.com/sports-illustrated-ai-generated-writers

NVIDIA CEO Huang urges faster AI development—to make it safer

https://www.goldmansachs.com/intelligence/pages/nvidia-ceo-huang-urges-faster-ai-development-to-make-it-safer.html

Technical/Dev/Chips

These ex-Apple employees are bringing AI to the desktop

https://www.theverge.com/2023/11/29/23981802/software-applications-inc-workflow-shortcuts-apple-employees-startup

This is a powerful harbinger that AGI might be the end-game.  I’ve always assumed specialists would beat generalists.  “A well-prompted frontier generalist AI (GPT-4) beats highly fine-tuned specialized LLM’s (Med-PaLM 2) in its area of specialty. Models are improving fast, though usual testing caveats.

https://t.co/4DBlB8OuLt

Silicon Volley: Designers Tap Generative AI for a Chip Assist.  Semiconductor engineers show how a specialized industry can customize large language models to gain an edge using NVIDIA NeMo.

https://blogs.nvidia.com/blog/llm-semiconductors-chip-nemo/

The Q* hypothesis: Tree-of-thoughts reasoning, process reward models, and supercharging synthetic data

https://www.interconnects.ai/p/q-star

Anthropic slashes AI pricing amid rising competition

Anthropic slashes AI pricing amid rising competition

Voltage Park launches massive new cloud for AI development

https://blog.voltagepark.com/voltage-park-launches-massive-new-cloud-for-affordable-ai-development/

Nvidia sold half a million H100 AI GPUs in Q3 thanks to Meta, Facebook — lead times stretch up to 52 weeks

https://www.tomshardware.com/tech-industry/nvidia-ai-and-hpc-gpu-sales-reportedly-approached-half-a-million-units-in-q3-thanks-to-meta-facebook

How Much Does it Cost to Use an LLM?

https://tomtunguz.com/gm-saas/

Exploring the Growing Convergence Between Blockchain and AI

https://pages.casperlabs.io/report/convergence-blockchain-ai

God Help Us, Let’s Try To Understand AI Monosemanticity

https://www.astralcodexten.com/p/god-help-us-lets-try-to-understand

Towards Monosemanticity: Decomposing Language Models With Dictionary Learning

https://transformer-circuits.pub/2023/monosemantic-features/index.html

Reshaping the tree: rebuilding organizations for AI

https://www.oneusefulthing.org/p/reshaping-the-tree-rebuilding-organizations

Stability AI Explores Sale as Investor Urges CEO to Resign

https://www.bloomberg.com/news/articles/2023-11-29/stability-ai-has-explored-sale-as-investor-urges-ceo-to-resign

We present MEDITRON, a set of new open-access #LLMs (70B & 7B) adapted to the medical domain, achieving new SoTA open-source performance on common medical benchmarks, outperforming GPT-3.5 and Med-PaLM,  and coming within 5% of GPT4 (counterpoint to the generalist link above)

Microsoft Chief Scientific Officer Eric Horvitz explains how new prompting strategies can enable generalist large language models like GPT-4 to achieve exceptional expertise in specific domains, like medicine, and outperform fine-tuned specialist models:

The Power of Prompting

Augmenting Human Intelligence – the IBM Point of View

Potential uses include- Fraud detection, personalized learning, personalized finance, early disease detection, patient discharge and bed management

https://www.ibm.com/downloads/cas/2ZDOY697

Building and Evaluating Advanced RAG Applications (free course)

https://www.deeplearning.ai/short-courses/building-evaluating-advanced-rag/

Exciting news from Broadcom! Today we’re introducing the industry’s first on-chip neural network with the Trident 5-X12, which doubles bandwidth and reduces power by 25%.

https://www.broadcom.com/products/ethernet-connectivity/switching/strataxgs/bcm78800

Starling-7B model, that dropped today, performs almost as well as GPT-4 in everything except math, reasoning and code!

NVIDIA has just released three new GPU-based Amazon EC2 Instances, offering higher performance and better scalability for AI workloads.

https://aws.amazon.com/blogs/machine-learning/introducing-three-new-nvidia-gpu-based-amazon-ec2-instances/

Education/Learning

New short course on sophisticated RAG (Retrieval Augmented Generation) techniques

https://www.deeplearning.ai/short-courses/building-evaluating-advanced-rag/

AI Summary of last week’s Intro to LLMs video

I created a notes version of Karpathy’s  “Intro to LLM” video (~5 min read)

Microsoft launched a free course for beginners to learn AI. 

The 12-week, 24 lesson curriculum is all publicly available on GitHub

Self-Driving Cars

“Thank you Waymo riders for taking over 700,000 trips this year with no human behind the wheel.”

Tesla starts releasing to employees FSD v12 – a critical update to self-driving effort

https://electrek.co/2023/11/24/tesla-releasing-employees-fsd-v12-a-critical-update-self-driving-effort/

China

Qwen

Qwen-72B has been trained on high-quality data consisting of 3T tokens, boasting a larger parameter scale and more training data to achieve a comprehensive performance upgrade. Additionally, we have expanded the context window length to 32K and enhanced the system prompt capability, allowing users to customize their own AI assistant with just a single prompt.

Qwen-1.8B is our additional gift to the research community, striking a balance between maintaining essential functionalities and maximizing efficiency, generating 2K-length text content with just 3GB of GPU memory.

Together AI

Open source AI platform raised $102.5M

https://www.together.ai/blog/series-a

Amazon

Really excited to share with customers Amazon Q—a new type of generative AI-powered assistant that is specifically for work and can be tailored to your business.

https://www.aboutamazon.com/news/aws-reinvent-2023-live?p=amazon-q

Amazon Q (Preview)

Your generative AI–powered assistant designed for work that can be tailored to your business

https://aws.amazon.com/q/

AWS Announces Amazon Q to Reimagine the Future of Work

https://press.aboutamazon.com/2023/11/aws-announces-amazon-q-to-reimagine-the-future-of-work

AWS re:Invent: Everything Amazon’s announced, from new AI tools to LLM updates and more

https://techcrunch.com/2023/12/01/heres-everything-aws-reinvent-2023-so-far/

Amazon’s AWS unveiled 2 new AI chips for use in data centers: 

Trainium2, the 2nd generation of its AI training chip, which AWS said is 4 times faster and 2X more energy efficient than its predecessor.  Graviton4, the 4th generation of its CPU, said to be 30% faster than its predecessor.

https://www.reuters.com/technology/amazon-announces-new-data-center-chip-microsoft-rivalry-intensifies-2023-11-28/

Amazon’s Q has ‘severe hallucinations’ and leaks confidential data in public preview, employees warn

https://www.platformer.news/p/amazons-q-has-severe-hallucinations

5 ways Amazon is using AI to improve your holiday shopping and deliver your package faster

https://www.aboutamazon.com/news/operations/amazon-uses-ai-to-improve-shopping

Amazon and Salesforce Expand Partnership to Add New AI Capabilities

Amazon and Salesforce Expand Partnership to Add New AI Capabilities

Business/Productivity

Intuit Adds Generative AI-Powered Tax Prep to TurboTax

Intuit Adds Generative AI-Powered Tax Prep to TurboTax

Rewind is a personalized AI powered by everything you’ve seen, said, or heard. Your colleagues will wonder how you do it all.

https://www.rewind.ai/

The 10-Year “Overnight” Success Story of Casetext (Legal AI)

https://www.ycombinator.com/blog/the-10-year-overnight-success-story-of-casetext/

Sales CRM

https://winn.ai/

NYT DealBook Summit | 2023: Elon Musk

Andrew Ross Sorkin has a remarkable conversation with Elon Musk

https://twitter.com/i/broadcasts/1MYxNoaVRjoKw

Anthropic

OpenAI rival Anthropic makes its Claude chatbot even more useful

Claude 2.1 doubles the number of tokens the chatbot can interpret to 200,000 and gains beta support for third-party APIs.

https://www.theverge.com/2023/11/21/23971070/anthropic-claude-2-1-openai-ai-chatbot-update-beta-tools

AGI

Prediction markets are putting AGI at 2031.

https://twitter.com/emollick/status/1729295140840202481/photo/1

Credits/Sources

Most of these links come from just a few incredible sources.  Please follow them:

Leave a Reply

Trending

Discover more from Ethan B. Holland

Subscribe now to keep reading and get access to the full archive.

Continue reading