Good morning. Autumn feels like it’s coming in earnest now — the mornings are crisp and cool, the leaves are changing and falling. I had my first cup of tea of the season.

And Elon Musk said that he’s going to launch a rocket to Mars in two years — in 20 years, he’ll start building a city on the red planet.

I don’t think there will be Autumn on Mars.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

• 🧠 AI for Good: Generating novel proteins

• 🏛️ Study: AI training is copyright infringement

• 📊 Venture funding is slowing down, but it’s more complicated with AI

 AI for Good: Novel Proteins with AlphaProteo

Google DeepMind last week unveiled AlphaProteo, an AI system that the company says is capable of designing novel, high-strength protein binders that can serve as the backbone of biological and medical research. 

The details: The process of developing protein binders that bind to a target molecule is lengthy and difficult. AlphaProteo, trained on protein data from the Protein Data Bank and more than 100 million predicted protein structures from Google’s other biological AI model, AlphaFold, has “learned” how molecules bind together. 

• “Given the structure of a target molecule and a set of preferred binding locations on that molecule, AlphaProteo generates a candidate protein that binds to the target at those locations,” Google said in a statement. 

• While not without its limitations, Google found that the model performed better than other design methods, indicating “that it could drastically reduce the time needed for initial experiments involving protein binders for a broad range of applications.”

Why it matters: Google said that the binders AlphaProteo is predicting can help boost advancements across a broad range of research, including in drug development, disease diagnosis and understanding and cell and tissue imaging. 

As the University of Washington’s Preetham Venkatesh said in a similar research paper last year: “There are many diseases that are difficult to treat today simply because it is so challenging to detect certain molecules in the body. As tools for diagnosis, designed proteins may offer a more cost-effective alternative to antibodies.”

Tech Workshop: Build a Voice AI Agent with Deepgram & Groq

Ready to write the code to build your own voice AI agent? Then this workshop is for you!

Our experts will guide you through each step and complex challenges in AI agent development like interruption handling, end-of-speech prediction, and low-latency function calling.

Plus, receive:

• $1,000 Deepgram API credits

• Access to a community of voice AI builders

Sign up before it's too late! Get 20% off with code: DV20

📆 September 20 @ 9AM-12PM PT

Study: AI training is copyright infringement 

AI companies have never shied away from the fact that they use copyrighted creative works to train their models; they’ve just claimed it’s fair for them to do so. OpenAI told the U.K.’s House of Lords in January that “it would be impossible to train today's leading AI models without using copyrighted materials."

But as the question of ‘fair use’ continues to be battled out in an ever-increasing list of court cases, a few technologists and legal scholars have begun to weigh in. 

What happened: A study commissioned by Germany’s Copyright Initiative in Spring — and conducted by a computer scientist and legal scholar — came to the conclusion that the process of training an AI model constitutes copyright infringement under European copyright law. 

• They found that infringement occurs at several stages: first, the collection, preparation and storage of copyrighted material, then the reproduction of that material ‘inside’ an AI model, then replication of parts of that material in generative output. 

• “As a closer look at the technology of generative AI models reveals, the training of such models is not a case of text and data mining. It is a case of copyright infringement –– no exception applies under German and European copyright law,” Professor Dr. Tim W. Dornis said. 

At the same time, Jacqueline Charlesworth, former general counsel of the U.S. Copyright Office, wrote in a recent paper that the ‘fair use’ defense does not apply to genAI training. 

She wrote that AI models aren’t human, cannot reason and “do not ‘know’ anything independently of the works on which they are trained, so their output is a function of the copied materials.”

“Large language models, or LLMs, are trained by breaking textual works down into small segments, or ‘tokens’ (typically individual words or parts of words) and converting the tokens into vectors — numerical representations of the tokens and where they appear in relation to other tokens in the text,” she said. “The training works thus do not disappear, as claimed, but are encoded, token by token, into the model and relied upon to generate output.”

Lock in a 6.9%* yield with a Bond Account on Public

On Public, there’s a simpler way to take advantage of the highest corporate bond yields since 2009. Meet the new Bond Account:

1. Invest your cash in a Bond Account

2. Lock in your 6.9%* yield

3. Earn interest payments 20x a year

Take advantage of historic yields

If you act now, you can lock in a 6.9%* yield before the Fed potentially cuts rates in September and beyond.

Turn your cash into bonds

With as little as $1,000, you can invest in a portfolio of ten investment-grade and high-yield bonds.

Get regular fixed-income payments

While you are invested in a Bond Account on Public, you'll receive regular interest payments.

Public is the all-in-one investing platform where you can build a multi-asset portfolio of stocks, options, bonds, crypto, and more.

See how you can lock in your 6.9%* yield

• Becoming ISO 42001 compliant shows your customers that you are taking the necessary steps to ensure responsible usage and development of AI, learn how with the ISO 42001 Compliance Checklist.*

• Want to become an AI consultant? 200+ students have already started and scored their first client in as little as 3 days. Request early access to The AI Consultancy Project.*

• Get high-quality meeting minutes, tasks, and decisions for all your online and offline meetings without awkward meeting bots. Save 10 hours every week and try Jamie now for free.*

• Did your car witness a crime? Bay Area cops may be coming for your Tesla (San Francisco Chronicle).

• California Passes Law Requiring Consent for AI Digital Replicas of Dead Performers (Variety).

• Musk denies report his xAI in talks over Tesla revenue (Reuters).

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

Venture funding is slowing down, but it’s more complicated with AI

The specter of AI in both the private and public markets has made for a pretty dichotomous saga over the past few months. 

On the public side of things, a strong earnings season from Big Tech did not provide the stock boost the sector was looking for, and Nvidia — that king of the AI boom — has been struggling since reporting earnings, a weight that has pulled down several other sizeable semiconductor firms. 

At the same time, amid recoveries to brief slumps across the major indices, market breadth has been steadily improving as investors begin to move away from the tech megacaps. 

But in the private sector — as Pitchbook put it in its August report — “investors cannot get enough of AI.”

A look at the scene: In August, total venture funding slowed down significantly, according to Crunchbase data, reaching only $18 billion. But AI companies represented the strongest sector. 

• The AI sector raked in around $4.3 billion in funding in August, 24% of August’s venture total. 

• Though it represents a significant come-down from a busy summer (AI firms raised $10.4 billion in May, $6.3 billion in June and $12.8 billion in July), the overall trend of AI funding remains strong.

Indeed, Crunchbase found that AI firms have set another record in the venture space; in 2024, some 35% of total U.S. startup investment went to the AI sector, the “highest percentage on record.”

In 2022, one in 10 dollars went to AI companies; in 2023, it was one in five. This year — so far — that number is one in four, according to Crunchbase.

• “Valuations have ramped up, and despite high prices, VCs have consistently been willing to invest,” Pitchbook wrote. 

• “There is a lot of hype around the technology, which many believe will be integral to reshaping the future. Their strong conviction has created a sense of urgency, as investors are consumed with the fear of missing out on big returns if they do not invest now.”

At the same time, Big Tech titans — from Microsoft to Amazon and Nvidia — have supercharged the private market by plowing billions of dollars (often in the form of strange deals and semi-acquisitions) into a range of AI startups. 

“With so much investor enthusiasm and available dry powder, AI valuations will likely remain elevated for a while,” Pitchbook said. “AI’s large price tag brings into question whether it can generate enough returns with such costly requirements … only time will tell whether these high valuations are justified.”

The backdrop to this — which we’ve discussed before — is AI’s $600 billion hole (the gap between revenue and incredibly high investment). For generative AI to generate (pun intended) the kind of returns that can justify that investment, these companies need to crack the enterprise with a clear, viable, reliable solution. 

But reports are mounting that this effort isn’t going well. 

In between reports that OpenAI is considering upping the cost of its subscription to ChatGPT, The Information reported that — of the small percentage of Microsoft customers who have adopted the company’s AI-powered copilot — some are pausing their use of the software due to annoying bugs and high costs.

The truth of the matter is that genAI isn’t the magical solution it has been hyped up to be.

Instead, it’s an unreliable tool with a high price tag that will likely need to get higher soon.

Unless these issues with reliability are solved — and there’s no evidence they will be — I can’t really imagine a widespread enterprise adoption of the tech that would be sizeable enough to prevent some sort of bubble burst, the first indications of which would be evidenced by Nvidia and the other semiconductors.

🤔 Your thought process:

Selected Image 1 (Left):

• “No one would put the downspout in the middle of the wall instead of at the corner.”

Selected Image 2 (Right):

• “I didn't think the waterfall in Image 1 could possibly be real.”

💭 A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.

Here’s your view on paying more $ for ChatGPT:

More than 41% of you said you wouldn’t ever pay anything for ChatpGPT; a third said you’d pay $20 a month and 15% said you’d pay $50. 12% said you’d pay whatever they charge and just one person would drop $100 a month on ChatGPT.

If you’re reading this, @ OpenAI, I wouldn’t go beyond $20.

$20/month:

• “$20 still seems high, but we've never paid for Google at the expense of dealing with "sponsored" links to weed through, so paying for an ad-free experience is worth the expense. Heaven forbid, the AI starts trying to sell us something!”

Absolute nothing:

• “I have found other solutions that are far superior.”

_{*Public disclosure: All investing involves the risk of loss, including loss of principal. Brokerage services for US-listed, registered securities, options and bonds in a self-directed account are offered by Public Investing, Inc., member FINRA & SIPC. Cryptocurrency trading services are offered by Bakkt Crypto Solutions, LLC (NMLS ID 1828849), which is licensed to engage in virtual currency business activity by the NYSDFS. Cryptocurrency is highly speculative, involves a high degree of risk, and has the potential for loss of the entire amount of an investment. Cryptocurrency holdings are not protected by the FDIC or SIPC.}

_{Alpha is an experiment brought to you by Public Holdings, Inc. (“Public”). Alpha is an AI research tool powered by GPT-4, a generative large language model. Alpha is experimental technology and may give inaccurate or inappropriate responses. Output from Alpha should not be construed as investment research or recommendations, and should not serve as the basis for any investment decision. All Alpha output is provided “as is.” Public makes no representations or warranties with respect to the accuracy, completeness, quality, timeliness, or any other characteristic of such output. Your use of Alpha output is at your sole risk. Please independently evaluate and verify the accuracy of any such output for your own use case.}