This week’s cover image depicts Robotic Russian nesting dolls to represent OpenAI’s GPT store announcement.

Executive Summary

A busy week with news coming out of multiple AI companies.  Here are the main stories, at an executive level, with full details in the body of this report.

  • OpenAI Dev Day: Tons of news coming out of the first Open AI developers day.  The two big takeaways are: a more powerful engine and the ability to create personalized AIs and share them in an app store.  GPT Vision can browse the web and shop on your behalf.
  • X Launches Grok: Grok is a competitor to ChatGPT launched by Twitter and is available for paying users
  • Amazon Olympus: Amazon is working on an AI that may be stronger than ChatGPT.  This could power Alexa and could potentially allow conversations and personal study guides to accompany its entire catalog of books and audio.
  • Bill Gates predicts: AI will upend the entire software industry
  • AI Email: Microsoft CoPilot integrates so well with Outlook email that we may see AI’s responding to AI’s soon. 
  • Wearable: A company called Humane has launched an AI pin with essentially no interface.
  • Creativity:
    • Everart.ai – consistent character generation is finally here
    • MidJourney’s beta site has launched
    • Motion Brush by Runway adds video control

Top Four Stories

Beyond the big tickets above, these are the links I’d click if you could only pick a few:

AI is about to completely change how we use computers and upend the software industry.
“You won’t have different apps for different tasks. You’ll simply tell your device, in everyday language, what you want to do… Agents are not only going to change how everyone interacts with computers. They’re going to upend the software industry.”
https://www.gatesnotes.com/AI-agents

Copilot for Outlook is very good and, as a result, is going to completely undermine how we all communicate with each other.
Microsoft Outlook can create professional drafts in a variety of tones and responses, with one-click buttons labeled with relevant suggestions. In many cases, AIs will be talking to AIs. Ethan Mollick, associate professor at Wharton, calls the effect on our psyche “the coming crisis of meaning”.
https://www.linkedin.com/posts/emollick_copilot-for-outlook-is-very-good-and-as-activity-7128047627664777216-LgJ0/

Communicators Cannot Control The AI Era
Jeremiah Owyang, partner at Blitzscaling Ventures, puts it bluntly, “Humans will rely on AI and the AI is going to get the information and make the decisions. That’s something communicators had control over, now [they’re] going to lose that control.” Owyang predicts AI will “end the abusive relationship we all have with search,” and SEO will “no longer be relevant”.
https://www.provokemedia.com/latest/article/provokeglobal-communicators-cannot-control-the-ai-era

Entertaining the AIs is getting harder. If they don’t like you they won’t distribute your posts to others. Puts pressure on to make better content.
https://twitter.com/Scobleizer/status/1723118375222591742

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!

OpenAI DevDay

Sam Altman Dev Day Opening Keynote

DevDay Overviews

GPT Leaks re Models Ahead

GPT-4 Turbo – The official new GPT-4

OpenAI has made a groundbreaking announcement with the release of the 128k context GPT-4 Turbo, now available at a third of the original price. This enhanced version of GPT-4 not only offers lower costs and increased speed but also boasts the capability to process prompts as extensive as an entire book. In a significant development for developers, the update includes a plethora of new features: the Assistants API for Code Interpreter and Retrieval, an updated knowledge cutoff for GPT-4 Turbo extending to April 2023, the introduction of the GPT-4 Turbo Vision API, Dall-E 3 API, and a highly realistic text-to-speech API with six preset voices. Additionally, OpenAI is set to release the new Whisper V3 model and has significantly reduced rate limits, further lowering development costs.  

You can now load the equivalent of 205 single spaced pages into the “working memory” of GPT-4. If you load less than 117 pages, the system can flawlessly recall details (at least in this test), after that, it has better recall of the very start and end of the text than middle.

Chart by Gregory Kamradt who has done some great exploratory analysis: https://lnkd.in/eECj5XNQ

https://www.linkedin.com/posts/emollick_you-can-now-load-the-equivalent-of-205-single-activity-7128577268381020160-vC9v

GPT-4 Turbo Summary:

1. 128K context window (300 pages of text)

2. More control; JSON Mode

3. Knowledge cut-off: April 2023

4. Multimodal; Whisper V3

5. GPT-4 fine-tuning; Custom models

6. Higher rate limits

Pressure Testing GPT-4-128K With Long Context Recall

In a detailed analysis testing GPT-4’s recall ability with up to 128K tokens, key findings revealed that recall performance diminishes beyond 73K tokens, with lower recall for facts placed at 7%-50% of the document’s depth. Facts at the document’s beginning are consistently recalled, indicating the importance of fact placement. The study utilized Paul Graham’s essays as background tokens, placing a random statement at various depths to assess recall, highlighting that less context typically results in higher accuracy. These insights underscore the lack of guarantees in fact retrieval with GPT-4, suggesting a strategic approach in context and fact positioning. The extensive testing, costing around $200 in API calls, indicates a need for further research to refine understanding of GPT-4’s capabilities.

GPT Vision with Web Access

The new OpenAI GPT-4 Vision model did my Christmas shopping in 30 seconds.

Custom GPT Builder

Chat Plugins Beta 

Learn how to build a plugin that allows ChatGPT to intelligently call your API. 

https://platform.openai.com/docs/plugins/introduction

Chat Completions – API Driven Consistency

Discover how to achieve reproducible outputs with OpenAI’s Chat Completions, ensuring consistent results across API calls  https://platform.openai.com/docs/guides/text-generation/reproducible-outputs https://cookbook.openai.com/examples/deterministic_outputs_with_the_seed_parameter 

Assistants API Beta

The Assistants API allows you to build AI assistants within your own applications. An Assistant has instructions and can leverage models, tools, and knowledge to respond to user queries. The Assistants API currently supports three types of tools: Code Interpreter, Retrieval, and Function calling. In the future, we plan to release more OpenAI-built tools, and allow you to provide your own tools on our platform.

https://platform.openai.com/docs/assistants/overview

Impact on Start-Ups

Tech startups using OpenAI’s APIs face new challenges and opportunities in the evolving AI landscape #aiebh https://www.businessinsider.com/openai-chatgpt-pdfs-ai-startups-wrappers-2023-10

Dynamic Maps for Travel Assistance

The AI assistant travel app Dynamic map pins based on a conversation along with basic integration of flight info… it’s simple but a harbinger of things to come.  Especially the death of the page view.  The more we chat with AI agents, the less we browse the web.

Training Set Partnerships

OpenAI seeks partnerships for building AI training data sets, aiming to steer and improve future AI model development  https://techcrunch.com/2023/11/09/openai-wants-to-work-with-organizations-to-build-new-ai-training-data-sets 

OpenAI’s GPTs Go Live After Fending Off Hackers

Hackers from the group Anonymous Sudan claimed responsibility for the OpenAI attack causing widespread outages in the week following DevDay.

https://gizmodo.com/chatgpt-blames-outages-ddos-hackers-gpt-rollout-stalled-1851006900

OpenAI Is A Lot More Vulnerable Than You Think 

All the press, money, and awards in the world won’t prevent OpenAI from the cold reality of competition. https://www.bigtechnology.com/p/openai-is-a-lot-more-vulnerable-than 

Amazon Olympus AI

Amazon is training a giant, 2 trillion parameters LLM codenamed “Olympus”…

  • They need a state of the art model to power Alexa.  
  • Having custom GPTs for every book, author and topic in the bookstore sounds like a no-brainer. 

Amazon dedicates team to train ambitious AI model codenamed ‘Olympus’ -sources

The model, codenamed as “Olympus”, has 2 trillion parameters, the people said, which could make it one of the largest models being trained. OpenAI’s GPT-4 model, one of the best models available, is reported to have one trillion parameters.https://www.reuters.com/technology/amazon-sets-new-team-trains-ambitious-ai-model-codenamed-olympus-sources-2023-11-08/

X Launches Grok AI

Overnight, Elon Musk and xAI revealed their ChatGPT competitor, Grōk.  Grok has real-time access to info on 𝕏.  This is a huge advantage over other models that can only access info based on the time they were trained.   Given that 𝕏 is the best place for breaking news, Grok could be huge for getting breaking news with clear and unbiased insight. 

-25,000 char. context window

-Screen refresh response speed

-“Live” search engine focuses on X for context

-API, Image recognition, and Audio recognition is planned

-A version of Grok will run “native” in Tesla using local compute power

Official Grok Announcement by X

https://x.ai/

Overview: Elon Musk debuts ‘Grok’ AI bot to rival ChatGPT, others

https://www.cnbc.com/2023/11/05/elon-musk-debuts-grok-ai-bot-to-rival-chatgpt-others-.html

xAI PromptIDE

Explore Grok-1 with xAI PromptIDE, a tool for researchers to utilize large language models with insightful analytics and complex prompting.

https://x.ai/prompt-ide

The one very powerful thing is Grōk AI has a “feel” for news like no other AI platform.

Humane AI Pin

Humane’s AI Pin: Revolutionizing Wearable Technology

  • Overview: Developed by former Apple innovators, Humane’s AI Pin introduces a new genre of AI-powered, screenless wearable devices. This fashionable accessory, designed to be pinned to clothing, aims to augment user capabilities while keeping them engaged in their environment.
  • Features:
    • Design: Standalone device, replacing traditional apps with streamlined voice-controlled workflows.
    • Display and Interaction: Features a touchpad and a Laser Ink Display for projecting information onto the palm. Controlled by voice and a “personic speaker” or Bluetooth headphones.
    • Privacy and Software: Privacy-centric, activating only on engagement, powered by a native AI OS and Ai Bus software for instant AI service connections.
  • Specifications and Pricing:
    • Includes camera, microphone, and depth/motion sensors.
    • Priced at $699, plus a $24 monthly T-Mobile plan.
  • Market Impact:
    • Funded with $230 million, valued at $850 million.
    • Showcased capabilities like real-time translation and health-related features.
    • Aims to make traditional smartphones and watches obsolete, contributing to the evolution of post-smartphone technology.

This is the Humane Ai Pin 

https://hu.ma.ne

Overview: All the details about Humane’s AI Pin, which costs $699 and has OpenAI integration.     

https://www.theverge.com/2023/11/8/23953022/humane-ai-pin-price-specs-leak

Silicon Valley’s Big, Bold Sci-Fi Bet on the Device That Comes After the Smartphone

Humane, a company started by two former Apple employees, says its new artificial intelligence pin can stop all the scrolling. Can it live up to the hype?

.

Another Overview of the Humane Pin

https://twitter.com/thealexker/status/1722679429971730720https://twitter.com/thealexker/status/1722680121847361596

Images/Creativity

Everart.ai 

Consistent character generation – still in beta but worth signing up

https://www.everart.ai/home

https://youtu.be/BH9fpZC8oWk?si=3XDqg_vzZZ8Aot44

MidJourney’s Beta Site has Launched

https://beta.midjourney.com/home

3D Imagery Tools

MotionGPT Beta Signup: https://www.deepmotion.com/motiongpt 

Animate3d: https://www.deepmotion.com/animate-3d 

Genie (Luma Labs): https://lumalabs.ai/genie 

Stable3d Signup: https://stability.ai/contact 

Mootion: https://www.mootion.com/landing 
All of these via Theoretically Media (my favorite channel for AI imagery news): https://www.youtube.com/watch?v=idS3NnV4dlg

Video 

Motion Brush by RunwayML for Gen-2 looks absolutely sick

Explore the 1stAI Machine, a revolutionary device for AI-generated video editing without traditional filming equipment  https://twitter.com/c_valenzuelab/status/1720487169997824163#aiebh https://vimeo.com/875098371

Alibaba Releases Video from Static Images

I2VGen-XL: High-Quality Image-to-Video Synthesis

via Cascaded Diffusion Models 

https://i2vgen-xl.github.io/

Health Care

Elon Musk’s Neuralink is working on a chip that will solve blindness

Abridge’s AI transforms patient-clinician conversations into EMR-integrated clinical notes, enhancing medical documentation  https://www.abridge.com/

Nabla Copilot’s AI generates clinical notes, streamlining healthcare documentation and improving patient care https://www.nabla.com/ 

Wysa’s AI conversations with millions globally shed light on mental health trends and the need for effective support models  https://www.wysa.com/ 

Youper offers an empathetic, clinically validated AI chatbot, supporting over two million people in mental healthcare https://www.youper.ai/ 

Glass Health’s AI platform aids clinicians in developing differential diagnoses and clinical plans, enhancing healthcare decisions  https://glass.health/ 

Discover how AI is revolutionizing diabetes detection through voice analysis https://www.medscape.com/viewarticle/998280 

Nuance’s DAX Copilot solution combines ambient AI with Microsoft Cloud, transforming clinical documentation https://www.nuance.com/healthcare/ambient-clinical-intelligence/explore-dax-for-clinicians.html

Good news! New AI tool predicts viral mutations before they happen, helping us stay ahead of pandemics.  This could help us design more effective vaccines and therapies early on in a pandemic.

Other Wearables

Wearable AI coach gives performance reviews to sales pitches

New AI noise-canceling headphone technology lets wearers pick which sounds they hear

Snapchat integrates ChatGPT for a new breed of interactive AI lenses  https://www.theverge.com/2023/11/9/23954372/snapchat-lenses-developers-beta-chatgpt-ai  

Google AI

Google Announces Expansion of AI Partnership with Anthropic

The two companies have worked closely since Anthropic’s founding in 2021, as the AI safety and research startup has built one of the largest Google Kubernetes Engine (GKE) clusters in the industry. Anthropic also currently uses AlloyDB, Google’s fully managed PostgreSQL-compatible database, to handle transactional data with high performance and reliability; and Google’s popular BigQuery data warehouse to analyze vast datasets and extract insights for its employees.

https://finance.yahoo.com/news/google-announces-expansion-ai-partnership-170000867.html

YouTube Comment topics summarized by AI

To help you easily understand and participate in comment conversations, we’re experimenting with AI that organizes large comment sections of long-form videos into easily digestible themes.

https://support.google.com/youtube/thread/242690316/testing-new-experimental-generative-ai-features

Google is bringing its AI-powered search to more than 120 new countries and territories

https://www.theverge.com/2023/11/8/23951134/google-search-generative-experience-sge-expansion-120-countries-territories

YouTube to test generative AI features, including a comments summarizer and conversational tool

https://techcrunch.com/2023/11/06/youtube-to-test-generative-ai-features-including-a-comments-summarizer-and-conversational-tool/

Ethics/Legal/Security

OpenAI will defend customers if they face copyright lawsuits. This is a big promise.

Elon Musk tells UK Prime Minister Rishi Sunak AI will put an end to work

https://www.bbc.com/news/uk-67302048.amp

Hacking Google Bard – From Prompt Injection to Data Exfiltration

https://embracethered.com/blog/posts/2023/google-bard-data-exfiltration

Andreessen Horowitz would like everyone to stop talking about AI’s copyright issues, please

https://www-businessinsider-com.cdn.ampproject.org/c/s/www.businessinsider.com/marc-andreessen-horowitz-ai-copyright-2023-11?amp

CEO & co-founder of Inflection AI  @mustafasuleyman explains how China will become a “really significant player” in AI over the next few decades.

We are not going to lose our lead to china any time soon 

https://www.bloomberg.com/news/articles/2023-11-09/us-will-lead-china-on-ai-for-years-deepmind-co-founder-says

Other AI Products

Entertainment recommendation engine (and pretty good!) Ask Likewise any entertainment question for instant answers, powered by AI. 

https://likewise.com/

Figma AI – Guided Whiteboarding Tools (diagrams, projects, etc)

At Figma, we believe AI should be more than a novelty—it should offer solutions to actual problems. With this in mind, we’ve integrated AI into FigJam to make it as useful as possible. Our mission? To help you instantly visualize ideas and plans, suggest best practices, and, of course, automate tedious tasks, so you can focus on the bigger picture.

https://www.figma.com/blog/introducing-ai-to-figjam/

An AI just negotiated a contract for the first time ever — and no human was involved. Luminance is the company.  Autopilot “handles the day-to-day negotiations, freeing up lawyers to use their creativity where it counts, and not be bogged down in this type of work.”

https://www.cnbc.com/2023/11/07/ai-negotiates-legal-contract-without-humans-involved-for-first-time.html

AR/VR/Gaming

Xbox collaborates with Inworld AI to develop advanced AI tools for game development https://venturebeat.com/games/xbox-partners-with-inworld-ai-to-build-ai-tools-for-game-development/ 

Stability AI’s latest tool uses AI to generate 3D models – Stable 3D, an AI-powered app that generates textured 3D objects for modeling and game development platforms like Blender, Maya, Unreal Engine and Unity.

https://techcrunch.com/2023/11/02/stability-ais-latest-tool-uses-ai-to-generate-3d-models/

Meta dropped a new NeRF paper. While it’s not the most compelling capture, notice the exposure changes as the VR headsets move around. HDR NeRFs just hit different. 

8-Minute Explainer on Gaussian Splatting

Technical News

NVIDIA’s Eos supercomputer just broke its own AI training benchmark record

The system can train a 175 billion parameter GPT-3 model in under four minutes. It needs just 7.2 seconds for BERT.  NVIDIA was quick to note that the 175 billion parameter version of GPT-3 used in the benchmarking is not the full-sized iteration of the model (neither was the Stable Diffusion model). The larger GPT-3 offers around 3.7 trillion parameters and is just flat out too big and unwieldy for use as a benchmarking test. For example, it’d take 18 months to train it on the older A100 system with 512 GPUs — though, Eos needs just eight days.

https://www.engadget.com/nvidias-eos-supercomputer-just-broke-its-own-ai-training-benchmark-record-170042546.html

Nvidia Is Piloting a Generative AI for Its Engineers ChipNeMo summarizes bug reports, gives advice, and writes design-tool scripts

https://spectrum.ieee.org/ai-for-engineering

UK invests $273 million in AI supercomputer as it seeks to compete with U.S., China 

https://www.cnbc.com/2023/11/01/uk-to-invest-273-million-in-turing-ai-supercomputer.html

Meta’s AI research group wants to keep releasing models for free despite criticism that Llama 2 isn’t open enough.

https://www.theverge.com/2023/10/30/23935587/meta-generative-ai-models-open-source

Hugging Face has a two-person team developing ChatGPT-like AI models

https://techcrunch.com/2023/11/08/hugging-face-has-a-two-person-team-developing-chatgpt-like-ai-models/

Microsoft has over a million paying Github Copilot users: CEO Nadella

https://www.zdnet.com/article/microsoft-has-over-a-million-paying-github-copilot-users-ceo-nadella/

Yi Open Source AI

https://www.01.ai/

European AI company Aleph Alpha raises $500M, competing in the global AI market  https://news.crunchbase.com/ai/openai-anthropic-competitor-germany-aleph-alpha/

GitHub of automated sports commentary

https://github.com/ggoonnzzaallo/llm_experiments/blob/main/narrator.ipynb

AI Agents

Announcing the new Lindy: the first platform letting you build a team of AI employees that work together to perform any task — 100x better, faster and cheaper than humans would. 

Watch how I build a Lindy to track competitors in 30s:

Here, I get a Competitive Intel Manager Lindy to spin up one Competitive Analyst Lindy for each of my competitors:

These “Societies of Lindies” can be of any arbitrary complexity. We even have a group of 4 Lindies building API integrations.

Lindies can work autonomously, and be “woken up” by triggers like a new email, a new ticket, a webhook being hit, etc… Here, I set up my Competitive Intel Manager Lindy to wake up every month and send me a new report.

https://www.lindy.ai/

Samsung AI

Samsung details ‘Galaxy AI’ and a feature that can translate phone calls in real time

https://www.theverge.com/2023/11/8/23953198/samsung-galaxy-ai-live-translate-call

Samsung AI Forum 2023 Day 2: Discussing Technological Trends and the Future of Generative AI

https://news.samsung.com/global/samsung-ai-forum-2023-day-2-discussing-technological-trends-and-the-future-of-generative-ai

Samsung unveils its generative AI model Samsung Gauss

Samsung Gauss consists of language, code, and image models and will be applied to the company’s various products in the future.

https://www.zdnet.com/article/samsung-unveils-its-generative-ai-model-samsung-gauss/

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