Share

Good Morning

I don’t usually nerd out on non-LLM machine learning announcements. But extraordinary claims have been made. I hope you are in a curious mood.

Now that agent swarm consumer products like Meta’s Muse are going mainstream, we are on the cusp of models that will make machine to machine (M2M) processing and communications faster and more efficient. In my humble opinion that in a nutshell is what Jev is.

Try Muse (in U.S. & Canada)

So what is this Jev AI exactly?

Jev has gone Semi-Viral as new kind of Model

About a week ago on September 15th a former OpenAI employee, Diogo Almeida, announced Jev:

Jev: A Probabilistic Classifier with a Manifesto

Jev is a classifier that generalizes very well. Almeida is credited with being one of the co-inventors of reinforcement learning from human feedback (RLHF) and thus a co-inventor of some of what which became ChatGPT. Jev has potentially the ability to make some aspects of agentic AI more cost efficient while it gets its name, Jev, after Jevons paradox.

We still know very little about Jev's internal architecture, training setup, and model size. TypeSafe.ai have raised $40 million Seed and built a product called Jev that is an AI model built for fast, structured decisions rather than open-ended text generation. So you feed it state + typed questions (Choice, Score, or yes/no), and it returns calibrated decisions in 70–500 ms at almost zero cost ($0.042/M input tokens). You can read their manifesto here.

System-1 Might be useful for Agentic AI

Jev is Typesafe AI's new System-1 model that skips token-by-token generation for instant decisions and the internet the past week have found all kinds of use cases for the technology. According to Nicolas Bustamante what we might be witnessing in AI is - err, well, a textbook example of Clayton Christensen’s Innovator’s Dilemma playing out in AI. We are at that weird stage where the market is commoditizing the bottom of the ML market: think classifiers, routers, scoring and context triage. So is Jev viral AI non-sense or something useful? I think that remains to be seen.

Almeida left OpenAI two years ago to start TypeSafe AI and release a transformer based model that is not an LLM that produces probabilistic calibrated decisions. That is, Jev is what Typesafe calls a “System 1” model: it outputs decisions directly, as probabilities and confidence scores, rather than autoregressively predicting the next token. It’s certainly making debate about advances in classifiers sound more sexy.

It’s a model that is garnering significant interest from developers. That is, Jev appears most useful for software automation. It might turn out that consensus among software developers will see it as a cheaper and more robust way to incorporate intelligence into their code. In a world where agentic swarms are using more and more of the total demand for inference compute, that kind of makes sense. This a generalist model that might be useful for swarms of agents to become even faster and less costly.

Almeida claims Jev is built for the decisions inside software. Now Jev has been released in early access and its stimulating the open-source community. Typesafe’s selling point is that Jev is up to 100 times faster and 100 times cheaper than conventional LLMs for certain tasks.

Read the Blog

The Era of Hundreds of Auxiliary Specialized kinds of Models is Coming

In today’s world developers can embed Jev in products for high-volume, low-latency decisions such as classifying/routing support tickets, scoring leads or content, safety/flagging checks, verification steps, agent routing, or other “if-statement–like” logic inside automated pipelines with likely hundreds of potential use cases that have at least some business value. Jev is a sign of things to come in probabilistic AI and one one of the many new kinds of (often open-source) models that can make products better.

In an AI boom it’s interesting what becomes a potential product. Classification, scoring, routing, and intent detection are not new machine learning problems. What TypeSafe AI is doing differently is building a model specifically around these kinds of typed, probabilistic decisions, rather than taking a general-purpose LLM and prompting it to behave like a classifier.

Typesafe AI claims that at $42 per billion input tokens and free output, intelligence becomes a resource accessible to all.

Read Jev Docs

A week after launch, Jev has already been added by popular model gateway platforms such as Vercel, Cloudflare, LangChain, and Langfuse to their stacks. Jev is like a weird viral decision-making AI in September, 2026 and yesterday they announced that Jev is now available to everyone, without a waitlist, with access starting at $5 in credits, equivalent to roughly 120 million tokens. This is not a sponsor, but you can create your account here. Jev is also anticipated to be useful in agentic AI tasks like model routing.

Jev can use Far Less Tokens than Incumbent LLMs for some Tasks

Jev is very similar in spirit to zero-shot classification and seem to be able to do things a bit better and a lot more cheaply. I’m super interested in how Jev could make agentic AI faster, smarter and more efficient. In that fundamentally Jev is an AI model optimized for “machine-consumable decisions” rather than human-readable conversation and therefore likely better positioned as infrastructure for software rather than a chatbot/LLM. Jev is closer to the essence of what machine learning is but powered by the transformer architecture and perhaps made more powerful in an era where LLMs are so prominent.

Jev is described as non-autoregressive and might be based off of the work or related to the work of others. This work was done previously in part by Nandakishor Mukkunnoth and which he explains in a LinkedIn post. There have also in the past week been many spin-off versions of Jev that have gone open-source or been announced. He also shared his story on Hacker News here. Jev doesn’t reduce hallincations per se but delegates them to the user in a sense. Anything that reduces token usage is bound to go viral more easily in the current climate of AI compute.

TypeSafe AI has re-designed zero shot classification substantially since the time it was popular (2019-20) and added multiple structured decisions, probabilities, parallel inference, and a new calibration-focused training approach all in terms that are fairly familiar for NLP engineers. Six years later saying it’s a new kind of model isn’t 100% correct, but it’s certainly re-packaged in ways that are relevant to the era of Agentic AI and the challenges developers and product managers face.

Jev Reminds us that LLMs are not all We Need

TypeSafe AI attributes Jev's efficiency to its specialized architecture, parallel sampler, and calibration-focused training. It’s a generalized model for much narrower tasks designed to make structured decisions directly and in parallel. Immediately it can help AI agents decide which tool to call next and whether to retry the task, by answering those questions in under half a second for $0.042 per million input tokens. Output is free. It makes you wonder how in the near future different kinds of transformer models will combine, complement and improve what LLMs are capable of doing today. What will the decisions trees and probability matrices of the future even look like? Jev really unlocks the imagination of what will become possible.

“You can break a workflow into individual judgments, run them in parallel, and combine them however you want.”

Jev is an architecture is optimized specifically for fast, structured decision-making and will radically reduce the cost of some layers of the product for developers all over the world. Think about what that means? Jev could arrive just at the time to become an important part of the decision-making layer underneath a growing ecosystem of AI agents.

The model is named after William Stanley Jevons, the 19th-century economist whose eponymous paradox describes how the falling cost of a commodity can lead to it being used more and more. Jevons paradox has indeed been that whisper witty water-cooler response moment on repeat all during the AI boom.

Models such as Meta's facebook/bart-large-mnli have long been used for lightweight zero-shot classification. But the developer tool kits keep getting refined.

“The problem is we are optimising for human language. We have been super good at human language for four years, but it’s not useful for automation because computers speak a different language.” - Diego Almeida, TypeSafe AI

Jev is most likely built upon an open-weight LLM, i.e. most likely a Chinese model. Almeida says Jev is trained exclusively on synthetic data using a technique he calls “reinforcement learning from calibrated decisions.” TypeSafe AI reports Jev at around 68% (accuracy) on its own workflow evaluation - which is to say we have no idea how accurate these results are on real-world tasks. Jev’s general accuracy is very much now being debated and time will tell.

News Models are Boosting what Developers can do

I think Jev is super interesting especially for developers who are builders working on real-time systems, games, agent loops, or anything that currently avoids LLMs because of latency or cost. I’ve seen an impressive amount of use cases online in just the past week being mentioned.

RLCD stands for Reinforcement Learning for Calibrated Decisions, the training method TypeSafe AI says it developed for Jev. In essence this makes classifiers more useful. Think about it RLDC here refers to training methods in which a computer program not only learns to make good decisions, but also to realistically assess how confident it is in doing so. The goal is a system whose self-confidence matches its actual hit rate.

A standard RL agent might assign a 98% probability or preference to an action that only succeeds 60% of the time. Here they are trying to correct this and for narrower tasks it could make sense. If Jev’s accuracy and confidence improves it could end up being a rather useful model. The online speculation around Jev’s architecture (this blog by Archer Hume) is also fairly fascinating.

“What I like about Jev: for years, large generalist models sucked almost all the oxygen out of AI.

But there are huge opportunities in much more specialized and customized models, built for specific tasks and languages and as a result, orders of magnitude cheaper, faster and more optimized. There are 3 million of them publicly available on HF” - Co-Founder of Hugging Face.

I believe Jev is just a beginning for new kinds of non-LLM models that are made for more specialized workflows, tasks and use cases. Many of these specialized models are trained on open-weight Chinese models and you can see how open-source AI is stimulating complementary machine learning and models to complement and make up for some of the limitations intrinsic to language based models, or LLMS.

Not only are LLMs compounding with compute but more specialized models are doing important work in industry domains where LLMs are not optimized to solve their key domain problems.

Subscribe now

Meta’s Muse has better First Two Weeks than ChatGPT?

According to Apptopia, Muse’s app on mobile devices has been downloaded more times during its first 12 days on the market than ChatGPT was in the 12 days after its debut. This is with regards to where Muse is now available, in the U.S. and Canada.

• In this subset of the data, Muse has now seen 1.8 million downloads to ChatGPT’s 1.3 million.

• Overall, Muse has seen 2.8 million total installs globally in its first 12 days.

App Figures made the calculation before that Muse had overtaken ChatGPT as the most viral new AI product in history (see chart below). Apptopia’s data also indicates that Muse’s U.S. daily active users are now higher than they were for ChatGPT at the same point after its launch.

To be fair we have to note that Muse is available across both iOS and Android, while ChatGPT launched on iOS only at the very beginning. Meta also has multiple apps with billions of users (to upsell Muse) and a larger marketing spend available to them.

Conflicted reports on X about the Jev Announcement

Thread about Potential New Use Cases for Jev

Will Jev survive as a Real Product?

Pre GPT ML Engineers are Confused

The Usual Copy-Cat Cadence from a Viral Announcement

Jev asked us to think differently about Old Problems

Confidence scores for business development outreach?

On Using AI as a Tool

Could this unlock AI on the Edge and on Mobile?

Jev can Generate the Best Option

Could Jev Accelerate Automated Advertising?

How about HR and Recruiting?

Could Jev accelerate Computer Use and take it to the next level?

Subscribe now

How about Design?

And classifying Research papers?

Share

As of today September 22nd, Typesafe AI’s Jev has basically been overwhelmed with signups, they’ve literally had to pause them. Suffice to say it was a successful launch so far.

Read more

Most pieces go out at 05:30 am EST (usually). AI Supremacy is a Newsletter about AI at the intersection of business, society, tech, human culture and the future of human civilization.

Listen to today’s guest piece:

Good Afternoon,

I didn’t plan on writing a piece today. But a few things happend that I need to talk about. Anthropic Claude Cowork’s plugins, especially for the legal industry has sparked a significant selloff in Software and (legal) analytics related stocks. The new plugin automates contract review, non-disclosure agreement triage, compliance workflows, legal briefings and templated responses, according to the company.

Speed up Legal Contract Review

The Capex is getting wild

Meanwhile Alphabet just reported earnings. With Google’s market cap ballooning over $4 Tr. - it’s gotten a bit out of hand. Now Google is saying that it will spend $185 billion CapEx guide for 2026. Alphabet (Google’s parent company) spent approximately $91.45 billion in capital expenditures (capex).

Alphabet’s Earnings are a tell of where AI is heading. As you can see that’s almost double what it spent in 2025. The reality will be even higher by the end of 2026. App Economy broke down some of the numbers:

• Google Cloud surged 48% Y/Y ($70B run rate). But the number to watch is the $240B backlog, up 55% sequentially.

• With the Gemini app hitting 750M monthly users, Alphabet is aggressively moving toward Agentic Commerce. Chrome’s new Auto Browse and the Universal Commerce Protocol (UCP) are maturing.

• With a fresh $16B round and a $126B valuation, Waymo is no longer a moonshot. It’s a market leader scaling toward a global infrastructure play. 15 million rides in 2025 was just the warm-up with 20+ cities expected in 2026.

Alphabet’s full-stack advantage in AI appears to be widening as OpenAI’s ecosystem building seems to be slow and full of missteps like the gigantic failure that is the Sora app.

Rivalry between Anthropic and OpenAI Heats up 🔥

In a surprising turn of events, Anthropic has decided to take aim at OpenAI’s Advertising push in Super Bowl commercial.

Sam Altman had said that Ads would be a “last resort” but seeing the success of Anthropic over their products, OpenAI appears to have gotten desperate. Anthropic have released several humorous videos showing how inappropriate such ads would look in a live chat. The company also confirmed that it has no plans to add ads to Claude.

Claude a Space to Think

Watch Videos

The videos attempt to depict how deceptive and ridiculous to the human experience Ads might be in chatbot like interfaces and queries. (Although I do find ChatGPT already sounds like that!)

Anthropic said the personal nature of users’ conversations with Claude would make ads feel “incongruous.” The Comms and PR team behind OpenAI, who presumably run Sam Altman’s X account also began to say some weird stuff on his behalf:

Read the Tweet

But in a week where it’s clear Codex is getting outcompeted and Claude Cowork plugins are leading to a stock selloff of Software companies that feels like a Software Apocalypse due to AI, Anthropic is getting the press that OpenAI used to get. And the evidence is fairly shocking:

Claude Code / Cowork are Overtaking OpenAI’s Codex

Sam Altman trying to do Ads about Cowork on his X account also show you the degree of desperation in his PR/Comms team. But reality is what it is….

Anthropic’s Revenue is expected to Surge in the late 2020s

Anthropic’s branding is also getting more sophisticated, I really like this Ad they did on YouTube a while back (about 4 months ago):

Alphabet Demonstrates Incumbents (BigTech) Getting Stronger with Gen AI

Google’s AI supremacy in LLMs and their integration in sprawling ecosystem advantages and distribution is showing AI is going to be a winner-takes-all market. Anthropic lost a huge deal with Apple for asking for too much money which Google subsequently won.

Ilia Karelin’s Top recent pieces:

Featured articles by today’s guest contributor:

• Commands vs Skills vs Agents in Claude Code: What Nobody Explains

• Official Claude Prompting: Everything You Need To Know

• I Stopped Scrolling for AI News: How I use Perplexity and Grok to filter hundreds of posts down to what matters

• AI Research 101: Learn Perplexity from Scratch In 1 Hour

It already feels like the end of the week, but we are not done. The labor market is showing some AI fatigue and a 2nm revolution in AI hardware is about to begin.

Claude’s AI Constitution is Controversial

Read more

Hey Everyone,

It’s not usually in my job description to be aware of what Microsoft, the world’s first, second most valuable corporation is up to in AI. Yet Microsoft is doing a lot of important thing…

Read more

Hey Everyone,

All eyes are on Nvidia Earnings. The anticipation among investors and analysts is building.

Something incredible happened three days ago, AMD announced it will acquire server builder ZT…

Read more

Hello Everyone,

I’m currently on my summer break, so this will only be a short note. But suffice to say I’ve been writing about Voice AI for more than a decade. I’m bullish on this overall trend now:…

Read more

Hey Everyone,

When I glance over at my articles on OpenAI in recent times, I have to admit it’s not looking good. The “generational AI” startup approaching $3.5 Billion in revenue, led by Sam Altman …

Read more

Hello Everyone,

You know there’s moments when as a consumer or writer, I do get a bit outraged. Full disclosure, this is one of those moments!

How is your virtual digital photographic memory? Would you be willing to lease this out to an AI? What about a Copilot, specifically, Microsoft Copilot Recall? What if such features were “on by default”. What does that even mean?

Total Recall by Microsoft

Microsoft wants us to feel comfortable with surveillance and data-scraping architecture, very comfortable, so much in fact that laptop users have eternal memory uplifted by artificial intelligence chatbots that will remember everything you’re doing on your computer and help figure out what you want to do next!

In a former life I used to be a tech privacy advocate, but in the era of Generative AI , privacy as we once knew it, seems to be a bit of an afterthought. Only Apple is likely to take privacy as a feature important, we’ll find out on June 10th. Let’s here mention that Microsoft has a long track record of security breaches, but the past few years have been particularly bad for the company.

I’m not a willing participant in Microsoft’s Copilot era currently and who in their right mind would opt-in to Windows 11 again? Would you trust Microsoft with total Recall of your data?

I’ve been a bit vague, but we’re not talking about a hypothetical product here, it actually has been announced, someone at Microsoft thought this would be a good idea:

Microsoft’s AI Recall explained:

👁️‍🗨️ Takes screenshots of everything you do in Windows 11 and uses local small language model to help you search through it

🔎 Stores screenshots locally. You adjust how long they are stored for and how much drive space to dedicate

🍍 Runs on-device NPU

⚖️ Will be on by default, you can turn it off. Or restrict it from snapping specific apps or websites.

Spyware in a Copilot 👁️

Gosh that’s like spyware in a Copilot (🤯) isn’t it? And a cheap clone of Limitless, formerly known as Rewind. You see, I’m actually interested in how Generative AI will supercharge Surveillance Capitalism. 👀

Doesn’t Windows already do this with keystrokes?

Microsoft's built-in keylogger is turned on by default for both Windows 10 and 11. You will need to manually turn the keylogger feature off in your device's Privacy Settings if you don't want Microsoft to log everything you type on your keyboard.

• How to disable Recall?

Let’s look at some of the videos of what Recall is.

What is Recall?

Recall takes snapshots of your screen and stores them in a timeline. Snapshots are taken every five seconds while content on the screen is different from the previous snapshot. Snapshots are locally stored and locally analyzed on your PC.

Source.

Via the Verge

Via the WSJ

Nadella thinks this is integrating “memory” into your work and PC life.

Screenshot from Rewind

What happens when the digital equivalent of the eye of sauron of a Software OS monopoly is the very Operating system you use by default exactly? Serious question though, should we feel comfortable with ‘memory enhancements’ on by default?

So What?

Microsoft's announcement of the new AI-powered Windows 11 Recall feature has sparked a lot of concern, with many thinking that it has created massive privacy risks and a new attack vector that threat actors can exploit to steal data.

The AI & Privacy Concerns

Featuring Sinead Bovell.

The Quick Take

Featuring the YouTube Creator BWOne. (Warren B.)

Microsoft continues to try to force features on their users. Bad idea for privacy, on by default sounds very creepy to the majority of people.

Leave a comment

Sharing is Caring

🔁 Every week I give free paid comps to my readers who share my work the most on Notes and elsewhere. That’s a $60-100 value.

Ask yourself and ask your boss, coworkers, partner and family. Do I want what equates to AI powered spyware literally baked into the very fabric of the operating system itself?

Share

Subscribe now

• Join our AI Chat on this issue to debate about Recall and share your opinions that is free for all readers.

• Read on for a deep dive on Microsoft Recall that is part Op-Ed and part research. Let’s get into it.

Read more

Hello Everyone,

Today I want to talk about AI and the stock market.

If you don’t think the stock market is in an AI bubble, pay some attention to Tech Earnings this week and the next. It’s April 25th, 2024 and it’s going to be a truly bizarre Q1 Earnings season.

While mentioning AI on Earnings' calls worked out well in the recent past, in my estimate, even BigTech are over-investing in Generative AI in 2024. Remember when Tesla claimed it was an AI company? Tesla's free cash flow was negative $2.5 billion in the quarter, a shocking 674% year-over-year drop. And yet, Elon Musk continues to try to sell us a Robotaxi dream and insist that Tesla is a technology company, and not just a EV car manufacturer with some bells and whistles. His promises better pan out, Tesla’s stock is already down 35% this year.

Meta’s stock that has been up a remarkable 137.7% in the past year, and while 98% of its revenue is from digital advertising, it also coincidentally thinks its an AI-Metaverse company. Meanwhile how is Meta’s R&D actually going? Meta’s Reality Labs unit, reported sales of $440 million for the first quarter and $3.85 billion in losses. The division’s cumulative losses since the end of 2020 have topped $45 billion. Now with Meta getting more aggressive with these Llama models, it’s also insisting it’s investing in Generative AI as a leader in the field.

What could possibly go wrong for the hyperscalers weaving AI dreams and unrealistic expectations for their shareholders?

Support and Dive Deeper

Support my work for as little as $2 a week. 🎓📚💡

Subscribe to Listen to this Article

• Many of my premium articles now include audio narrated versions for easy listening while you live your life.

Subscribe now

Read more

Image: “Shiba Inu” ( $SHIB )

Hey Everyone,

What is the relationship between Bitcoin, crypto’s ecosystem and AI? How are “shitcoins” involved? The term shitcoin refers to a cryptocurrency with little t…

Read more

Hello Everyone,

Nvidia’s GTC is finally here in 2024. Dubbed the Woodstock festival of AI by Bank of America analysts, GTC this year is set to draw 300,000 in-person and virtual attendees for the debut of Nvidia Corp.’s B100.

I’m really a fan of Nvidia, so I was really looking forward to their GTC conference which has started (March 18th to 21st). I’ve been waiting for this keynote address by Jensen Huang for weeks so here it is:

Keynote GTC Nvidia Conference March 18th

• Back when Nvidia was known for GPUs for gaming and crypto, GTC wasn’t such a big deal. But now that they are known for A.I. chips it feels like everything has changed. The NVIDIA GPU Technology Conference (GTC) is NVIDIA's premier event for AI innovators, developers and enthusiasts.

• As I have begun to dabble in Semiconductor and AI chip coverage, I’m overwhelmed with the dynamism of the datacenter and AI chip industry. Check out my Semiconductor Things, a new Newsletter.

Semiconductor Things™

This Newsletter was built to solve the pain point of getting the latest news on A.I. chips, semiconductors, datacenter innovation and chip news.

By Michael Spencer

I’ve been looking forward to more information on B100, and B200 (two of which can morph into a GB200) for a long while. New chip architecture promises big performance boosts that could reduce the compute of Generative AI models. Meanwhile Nvidia has a rare monopoly on the A.I. chip and datacenter space and will now experiment with custom AI chip orders.

The Stock Market Loves Generative AI

Nvidia’s stock is up 241% in the past year alone YOY. For a company that size, that’s unprecedented in the history of BigTech. There is a lot riding on Nvidia’s success including the valuation and ‘market sentiment’ of the entire Generative AI space.

The NASDAQ 100, a benchmark for Tech stocks, is up over 40% in the last year.

The GPUs that Nvidia produces and the H100 that had such a high demand is now getting older, and a new generation of AI chips and custom AI chips will be led once again by Nvidia, globally.

“Nvidia today accounts for more than 70 percent of A.I. chip sales and holds an even bigger position in training generative A.I. models, according to the research firm Omdia.”

Blackwell Platform for Nvidia in 2024

Powering a new era of computing, NVIDIA on March 18th, 2024 announced that the NVIDIA Blackwell platform has arrived — enabling organizations everywhere to build and run real-time generative AI on trillion-parameter large language models at up to 25x less cost and energy consumption than its predecessor. - Learn more.

• DGX Grace-Blackwell GB200: exceeding 1 exaflop compute in a single rack.

• Put numbers in perspective: the first DGX that Jensen delivered to OpenAI was only 0.17 Petaflops.

• GPT-4-1.8T parameters can finish training in 90 days on 2000 Blackwells.

• Some analysts are cleverly saying that a New Moore’s law is born.

In 2024, hardware and software, incumbents and new startups are converging in a unique way for the future of Generative AI, like we’ve rarely ever seen before in the history of technology. It’s powering so much innovation in the U.S. and China, it’s hard to keep track of. Certainly very few Western news outlets or Creators are aware of what’s going on in China.

Nvidia is in many ways, the touchpoint for the future of A.I. hardware. I cannot stop writing about Nvidia in recent months:

1. Why Nvidia and not OpenAI is Generative AI's Symbolic Company

2. How BigTech Invested in AI Companies in 2023

3. Nvidia's Market Cap Overtakes Google & Amazon in Early 2024

• To see my posts on Nvidia in chronological order go here.

Insights from GTC 2024

A shorter video on Blackwell (2:36):

According to the company, the Blackwell GPU architecture features six transformative technologies for accelerated computing, which will help unlock breakthroughs in data processing, engineering simulation, electronic design automation, computer-aided drug design, quantum computing and generative AI — all emerging industry opportunities for NVIDIA.

• Generative AI

• Quantum computing

• Accelerated computing

• Data processing

• Engineering simulation

• Electronic Design automation

• Computer-aided drug designs

Nvidia’s latest chip architecture promises to be impactful across technologies and R&D as a whole.

Nvidia also announced later Project GR00T, our new initiative to create a general-purpose foundation model for humanoid robot learning. Nvidia is also making strides in things like Robotics (learn more) and Quantum computing partnerships.

Subscribe now

Read more

Sora as an announcement of a text-to-video model has surprised a lot of people. Seeing is believing in this case:

Read more

Hey Everyone,

I’ve been covering AI news fanatically for around 26 months, or in the area of 104 weeks. Mid February, 2024 is the biggest week of A.I. announcement I’ve ever seen.

• Nvidia’s Earnings beat

• OpenAI’s text-to-video Sora

• Groq’s LPU Chip increases speed (latency) of responses substantially

• Google Gemma

• China’s Moonshot AI raises US $1 Billion

• SoftBank founder Masayoshi Son wants $100 billion for a new AI chip venture

• Mistral Next

• Amazon BASE TTS makers claim “Emergent Abilities”

• Foundational Model for Biology, Bioptimus Startup

• OpenAI building its own Search Engine

• India announces Reliance’s BharatGPT ‘Hanooman’

• Intel launches Systems Foundry for AI Era

• Stability AI releases Stable Cascade

• Intel signs Microsoft as major Foundry Partner

First 🥇 AI Supremacy Reader Survey

🗣 After two years of stellar growth I built a survey with my long-standing realization that I need to align my work with my readers better. If you have a few minutes to help me do that, here is a quick survey.

• Skip as many questions as you like, but your participation is crucial for me to build a better reading product customized to your preferences.

• I will be giving out free comps to participants at random, so you might win a free paid subscription for up to one year, an $80 value.

Start Survey

🥇 Top AI Rundowns 🏆

These sources are good to skim for seconds per day, when you just want a glance of major headlines.

There are a great many ways to keep in touch with A.I. News on a daily basis. Here are some of the top ones in Newsletter rundown formats:

1. Ben’s Bites

2. AI Valley

3. The Deep View

4. OpenTool’s

5. TLDR AI

6. AI Breakfast

7. The Rundown

8. Semafor Tech

9. The Neuron

10. AlphaSignal

Even if you just hit one of these a day randomly, you should be alright. I don’t recommend Rundowns on Substack, since they are not optimized for skimming. You don’t want to be spending more than 10-20 seconds in these flash reads. I see Substack as more for deep dives, essays, education and curation compilations.

Image and work by of The Strategy Deck.

ICYMI

Here are some of our recent articles:

• AI can tell the Sex of your Brain

• The full potential of AI in techno-optimism

• How the Apple Vision Pro Could Rewire Our Brains by Guest post

• OpenAI and Its Startup Fund's Venture Investments - Guest post

• Nvidia's Market Cap Overtakes Google & Amazon in Early 2024

• Can AI learn language like we do?

Share

New Sora videos just dropped on TikTok by OpenAI team and people are going crazy 🐶. Nothing in this video is real.

Subscribe now

Nvidia’s rise in terms of market cap reminds many of Cisco in the dot.com bubble. There’s no other valid comparison really. It’s nearly difficult to put into context what we are witnessing.

Given Nvidia’s earnings there’s no reason to believe they don’t hit that $2 Trillion market cap soon. Let’s try and break all of this down.

Read more

Hey Everyone,

The idea of AI agents do things for us is not new. I covered on this Newsletter the Rabbit r1, that I thought was a pretty innovative take on AI helping you do stuff with your apps with an affordable device with a nice form-factor. You actually get what you paid for. They sold out.

Now it appears, OpenAI wants to do this as well, recently upgrading ‘memory’ of past conversations it has had with you and importantly has been working on building similar AI agent tech to, wait for it, ‘use apps for you.’ Take over your device? Automatically? Really.

Sam Altman seems always to want to steal the PR headlines while OpenAI’s product timeline aggressively appears to want more of a person’s data without clear benefits. Google made similar promises years ago, that never manifested. Does anyone remember Google Duplex? Chatbot and AI agent trends have a way of going bust.

OpenAI wants to build a next-level personal assistant, all the while OpenAI’s CEO warning of “societal misalignment” as a risk. OpenAI always says one thing, and does another. OpenAI at least is a product focused AI startup, innovating just enough to sort of hold your attention, even as you being to use ChatGPT less and less.

• Perhaps harvesting more and more data from their users is not really aligned with the future of personal assistants. Microsoft’s pricey Copilots and OpenAI’s tactics are going to lead to buyer’s remorse I predict. And here is the simple reason.

• A majority of the American public is skeptical about artificial intelligence, and the companies advertising (like at the Super bowl or Sam Altman’s constant public appearances) are attempting to humanize the technology. According to a recent Pew Research Study, 52 percent of Americans are “more concerned than excited about the growth of AI.”

• In 2024, wages have not even adapted to the rise in inflation. The majority of people and consumers don’t find Generative AI very useful in their daily lives or even at their job.

• Only 10 percent of the public are more excited than concerned about AI’s possibilities. Hint, they are’t ChatGPT daily active users and don’t want a company associated with Microsoft to leverage their personal data - it’s not, aligned to what these customers (most people), actually want.

• Those AI bros, they are a minority. Which makes you realize just what an echo-bubble the folks on X or LinkedIn are who concern themselves with AI.

The prognosis is not good for the latest wave of AI agents. Society just isn’t ready.

Share

There will be a collective lack of consent to what OpenAI is trying to do. Let’s take a look at what they say they are trying to accomplish potentially with their AI agents even with the little information that has been strategically leaked.

How to Sign-up 📑

Discounts on Premium - to get Full Access

If before March, 2024:

❄️ Special Offer to Go Paid 🌨️🧊☕️⛄️

This special 2-year Anniversary offer expires after February, 2024. It is only available for yearly plans. This amounts to $53.60 a year, or $4.16 a month, or $0.26 an article.

Get 33% off forever

🔴 To redeem this offer, click on the button above here. ☝ You can also choose to pay the full regular price if you choose to be among the 500+ premium members who supports my work.

Read more

Hello Everyone,

So 2024 begins a period of human history where the AI-arms race and the regional superpower battle for A.I. Supremacy takes on distinctly Chinese characteristics.

I’ve always been fas…

Read more

Hey Everyone,

It feels like we’ve been waiting for Google Gemini for years, but it’s really just months. This was supposed to be the product to compete with GPT-4 and ChatGPT. But does it even do tha…

Read more

Image: Nvidia's HQ has been designed to enhance productivity.

Hey Everyone,

If you are going to summarize what has occured 2023 in A.I., Nvidia’s performance really is the center of where the action is.

Just to illustrate the surprising year we’ve had in Generative A.I. euphoria, I think the best reflection of this is actually Nvidia’s recent Q3 Earnings.

While OpenAI has 1 Billion or so revenue, it’s very far from being profitable, while Nvidia is now part of the magnificent seven, entering a unique pantheon of companies with unprecedented power in the future of technology.

While recently Sam Altman has been a huge distraction, Anthropic’s Claude 2.1 release and Nvidia’s earnings shows you just how misplaced that hype is. OpenAI or ChatGPT are from the center of the gold-rush of LLMs in reality.

Nvidia reported Tuesday November 21st, 2023 blowout third-quarter (Q3) results and offered stronger Q4 guidance as AI-led chip demand bolstered growth in its high-margin data center business. 

Nvidia’s stock before today’s open is up 249% for 2023 so far, which is unprecedented for a company worth over $1 Trillion dollars. Tech execs have voiced concern that the development of artificial intelligence is concentrated in the hands of too few companies and Nvidia’s sudden unworldly success is a case in point.

Jensen Huang, CEO of Nvidia’s personal wealth is a case in point. Worth roughly $14 billion at the start of 2023, Huang is now worth nearly $44 billion thanks to his ownership of many Nvidia shares, and he is now the world's 27th richest person.

Let’s break down some of Nvidia’s earnings numbers to get a better idea of what has transpired for this company in 2023.

Nvidia’s market cap is 1.234 T. (Trillion) as of late November, 2023, and the stock is nearly $500. Nobody could have predicted that at the beginning of the year or even last year. Nvidia is by far the most important company right now in A.I.

To get access to my best deep dives, become a paid subscriber.

Subscribe now

Read more

Hey Everyone,

I’m pretty interested in what happens to xAI, Elon Musk’s AI startup.

Elon Musk's artificial intelligence startup xAI will release its first AI model to a select group on Saturday. He w…

Read more

Hey Everyone,

This is a special Saturday edition of A.I. Supremacy, but something came up that just couldn’t wait. Metaphors like “robots coming for our jobs” surround the fear of automation. Many of my readers don’t see it happening.

🏆Sponsored⭐

AI: What Consumers Say

Will consumers reward ethical data use in the age of AI? We completed a first-of-its-kind national survey to examine sentiment around artificial intelligence and the personal data that powers it.

Download the Report

After the exposition by on General Purpose Robotics startups, I had some private conversations with my readers who were saying that humanoid robots won’t become useful soon.

While I might have agreed a few years ago, recent advances have changed things somewhat I believe, and I really wanted to explain how Amazon is testing humanoid robots out.

Large language models might also give robots new ways of learning, Eureka bridges the gap between high-level reasoning (coding) and low-level motor control.

Eureka, an open-ended agent that designs reward functions for robot dexterity at super-human level

Eureka, is a human-level reward design algorithm powered by LLMs. Eureka exploits the remarkable zero-shot generation, code-writing, and in-context improvement capabilities of state-of-the-art LLMs, such as GPT-4, to perform evolutionary optimization over reward code.

See Eureka on Github.

Read the Paper

On a benchmark suite of 29 tasks across 10 robots, Eureka rewards outperform expert human-written ones on 83% of the tasks by 52% improvement margin on average.

• Is pen spinning really a big deal?

The Robots are Coming

I can fully understand that most don’t believe human-shaped robots with dexterous hands will be staffing warehouses and retail stores, tending to the elderly and performing household chores within a decade or so like the startups are claiming, but we do know that big players like Walmart and Amazon are very interested in robots performing tasks and some of those will automate jobs previously done by humans.

Why would Amazon not acquire the company behind the robot Digit? Oregon-based start-up Agility Robotics with backing from Amazon has come a long ways in a short time. Amazon already invested a significant amount of capital.

Digit can walk forwards, backwards, and sideways, as well as crouch, and stands 5ft 9in (175cm) tall. It can carry up to 35lb (16kg).

How Quickly did Robots Accelerate at Amazon?

Amazon’s infatuation with robots isn’t exactly new:

10 years ago, robots were practically non-existent in their global warehouse and distribution network.

But take look at this acceleration…

2013: 1,000 robots
2014: 15,000
2017: 100,000
2019: 200,000
2021: 350,000
2022: 520,000
2023: 750,000

Subscribe now

Clearly in the 2030s Amazon will be working with millions or robots.

Read more

Hey Everyone,

In an era of fanciful Generative A.I., large language models and robotics, even Disney is living up to its reputation.

Is this a precursor to social robots? If you think about it, Disney has designed with this robot those specific motions that convey emotions, character, and personality, using all the little designs ‘tricks’ that have made Disney successful in animation for decades. But if this is magic, how will the social robots of the future deceive us?

In machine learning terms, this means the robot is a triumph of Adversarial Motion Prior (AMP): a technique that learns the human preference by training a classifier on what we consider "emotional & cute". How far will the consumer robots of the future hack our tendency to anthropomorphize everything, even in a world that has been stripped of its humanity and tribal intimacy.

According to Nvidia researcher Jim Fan, AMP in GAN literature, this is called a discriminator. Disney artists are good at creating such a dataset. You can then add AMP as an auxiliary reward in simulation to nudge the robot towards desired behaviors.

According to Meta’s ‘Group Director’ Martin Harbech, with advancements like this, it’s not hard to imagine a not-too-distant future where robot companions with unique personalities are able to assist us in a range of tasks - from fun and play to homework and knowledge. One wonders what is the right interface for GPT-4 like technology to arrive for the masses?

Ghost in the Shell of Simulated Artificial Empathy

Is it an A.I. device or more properly the robots that will make up our environment in the not-so-distant future? An ambient future where A.I. is everywhere and is more than just large language models in apps, awkward chatbots, ‘augmented’ products and ever more capable or intelligent software. Disney’s robot makes us wonder about how we will relate to the ghost in the shell of the machines of the machine age.

What happens when robots and LLM entities develop their own uniqueness? What is the limit of the character in the A.I.? Will that imbue them with soul? Will so-called compassion robots earn the privilege of having rights? Character.AI, the a16z-backed AI chatbot startup from ex-Google AI researcher wants us to converse with chatbots in groups, Group Chat feature, as it’s called, allowing users to create a group chat with their favorite AI characters only or it can feature a mix of both humans and AI companions, the company says.

It begs a lot of questions. What happens to human intimacy in a symphony of chatbots, AIs, robots with animations and so-called AGI around every corner?

The Robot with an “IT-Factor”

Disney’s Wall-E like robot has no official name, but combines procedural animation, modular hardware, and reinforcement learning to be able to design and program a walking character capable of these unique gaits and traits.

Subscribe now

Read more

Hey Everyone,

We are going to cover the State of A.I. Report tomorrow, but today I wanted to think more deeply about AMD.

To understand the Generative A.I. wave of 2023 it’s helpful to understand Nvidia’s dominance in A.I. chips and its own investment in A.I. circa 2017, or around five years earlier. In 2023, Nvidia has also become an active funder of Generative A.I. startups as well.

AI Chip Market Set to 10x in just 7 Years

Driven by adoption of artificial intelligence (AI) across multiple industries, the global AI chip market generated nearly $29 billion in 2022, according to a report from Next Move Strategy Consulting. The market is expected to generate nearly $305 billion by the end of the decade, boasting a compound annual growth rate (CAGR) of 29% from 2023 to 2030.

As Nvidia has grown stronger in this area, the harder it has become for companies attempting to build competing chips. I am very interested in AMD’s attempt to catch up with Nvidia for LLM training. Back in July, 2023 MosaicML, that was acquired by DataBricks for $1.3B, published some interesting benchmarks for training LLMs on the AMD MI250 GPU, and said it is ~80% as fast as an NVIDIA A100.

In addition to developing their own AI chips in Athena, Microsoft it also came out in May, 2023 is helping to fund AMD’s expansion into AI Chips. About two weeks ago it came out that Machine learning startup Lamini revealed its large language model (LLM) refining platform was running "exclusively" on The House of Zen's silicon. Lamini claims (The Register) its platform, which has attracted interest from Amazon, Walmart, eBay, GitLab, and Adobe, to name a few, has been running on "more than a 100 AMD GPUs in production all year" and could be scaled up to "thousands of MI GPUs."

Subscribe now

Read more