ICYMI: Top Headlines from the Week

1. 🖥️ Anthropic Joins The AI Agent Race, Says It Can Control Your PC

2. 🕵️‍♂️ Microsoft's First AI Agents Are Here for Team Scaling

3. 🎬 Runway Launches Act-One for Character Animation

4. 🖼️ Stable Diffusion 3.5: High-Fidelity Images on Consumer Hardware

5. 🛡️ Biden Administration Issues National Security Memorandum on AI

6. 🆕 Meta Unveils New New Research and Open Models

7. ⚖️ Character.AI Faces Wrongful Death Lawsuit Over Teen Suicide

8. 👨‍💻 Elon Musk's xAI Launches API Letting Third-Party Developers Build Atop Grok

9. ⚖️ News Corp, Dow Jones, and NY Post Sue Perplexity for Copyright Infringement

10. 📜 Ex Open-AI Researcher Publishes Essay Arguing OpenAI Shouldn't Be Protected By Fair Use Doctrine

❶

The Fair Use Doctrine — Still Relevant in The AI Era?

With lawsuits popping up every other day from publishers and newspapers against AI companies like OpenAI and Perplexity that produce AI search overviews based on web content published on those companies platforms, it feels like a good time to dig into what laws they are actually basing their cases on.

What is the Fair Use Doctrine?

The fair use doctrine is a legal principle in copyright law that allows limited use of copyrighted material without permission from the copyright holder for purposes such as criticism, commentary, news reporting, teaching, scholarship, or research.

Origins

Dating back to common law precedents in the 18th and 19th centuries, the Fair Use Doctrine was formally established with the Copyright Act of 1976.

Fair use is determined on a case-by-case basis by considering four factors:

• The purpose and character of the use (including whether it is commercial or nonprofit educational)

• The nature of the copyrighted work

• The amount and substantiality of the portion used

• The effect of the use on the potential market for or value of the copyrighted work

No single factor is determinative - all four must be weighed together.

Commercial uses can still potentially qualify as fair use, though they face greater scrutiny. Fair use aims to balance the rights of copyright holders with the public interest in the use of copyrighted works.

Key Updates

• 1994: Campbell v. Acuff-Rose Music Inc. — emphasized the importance of “transformative” use in fair use analysis.

• 2021: Google LLC v. Oracle America, Inc. — questioned whether APIs are copyrightable.

• 2023: Andy Warhol Foundation v. Goldsmith — the Supreme Court ruled in favor of Goldsmith (the photographer) whose photograph of Prince was used in a commercial work by Warhol, stating that it was not “enough” of a transformation.

What About Weird Al?

“Weird Al” Yankovic is an iconic American musician and parodist who has been creating humorous versions of popular songs for over four decades. Rising to fame in 1984 with “Eat It,” a spoof of Michael Jackson’s “Beat It,” Yankovic has since produced numerous hit parodies including “Smells Like Nirvana,” “Amish Paradise,” and “White & Nerdy.”

Though Weird Al’s parodies are near exact melodic copies of the other artists work, he’s protected under fair use. Why?

So…it’s about the profits.

After issuing a cease and desist letter last week, The New York Times accuses Perplexity of:

• Using its content without authorization

• Unjustly enriching itself by using the Times’ journalism without a license

• Circumventing the Times’ anti-scraping and anti-bot measures

News Corp subsidiaries Dow Jones & Company (publisher of The Wall Street Journal) and NYP Holdings (owner of the New York Post) have filed a lawsuit against Perplexity AI in the Southern District of New York.

Allegations include:

• Copyright Infringement: News Corp accuses Perplexity of “massive” unauthorized copying of their content to train its AI and build its search index.

• Trademark Violations: The lawsuit claims Perplexity uses News Corp trademarks without permission and attributes false information to their publications.

• “Freeriding”: News Corp alleges Perplexity’s business model involves exploiting their protected content without compensation.

In both conflicts, the core of the issue is that Perplexity’s use of the data is very likely to infringe on the publisher’s profits.

Basically, if you use Perplexity to search a topic, you simply read the summary to get the information you want, and going to the publishers website is no longer needed. Both their web traffic (which is mainly monetized through ad revenue) and subscriptions are at risk.

What’s The Solution?

If nothing is done, Perplexity may truly end up with a parasitic fate. If people effectively stop paying publishers (preferring AI overviews instead), they no longer have incentive to publish new information, and Perplexity will have worse and worse data to use in providing their answers.

We could potentially imagine a world where large publishers begin to fail, and smaller, more agile outfits / individual reporters rise up to compete for the smaller market. It might not be that terrible for AI companies, since gathering data from widespread sources isn’t much of a problem. But the quality of the data is still likely to decrease in that newer paradigm, since larger companies like the New York Times ensure more stable funding for their journalists and maintain robust quality control operations as part of their business.

Interestingly, News Corp already has a licensing deal with OpenAI, valued at $250 million over five years. There are plenty of these sort of deals popping up this year for OpenAI.

1. Axel Springer (December 13, 2023)

2. Le Monde (March 13, 2024)

3. Prisa Media (March 13, 2024)

4. Financial Times (April 29, 2024)

5. Dotdash Meredith (May 7, 2024)

6. News Corp (May 22, 2024)

7. The Atlantic (May 29, 2024)

8. Vox Media (May 29, 2024)

9. TIME Magazine (June 27, 2024)

10. Condé Nast (August 2024)

Google signed a single deal with Reddit in February 2024, valued at $60 million.

Anthropic has not signed any major deals.

Perplexity, on the other hand, launched a revenue sharing program called the Perplexity Publishers’ Program:

• Revenue Sharing Model: When Perplexity generates revenue from an interaction that references a publisher’s content, that publisher will receive a share of the revenue. This is based on advertising revenue, not a flat licensing fee.

• No Content Licensing: Perplexity maintains that it doesn’t need to license content because it’s not training its language model on publishers’ writing. Instead, it sources facts from journalism for its query responses, which the company claims falls under fair use.

• Initial Partners: The program launched with several major publishers, including Time, Fortune, Der Spiegel, Entrepreneur, Texas Tribune, and WordPress.

• Additional Benefits: Partners receive free access to Perplexity’s APIs, a one-year license to Enterprise Pro for all employees, and analytics through ScalePost.ai.

• Future Plans: Perplexity aims to onboard 30 publishers by the end of the year and is exploring collaborations with publishers’ advertising sales teams.

Looking Ahead

With that rev-share program, it sort of sounds like Perplexity wants to establish a social media style paradigm for news and information media, with itself as the platform where people go to find news and information, and publishers acting as content creators.

Will it work? Probably not, at least not while the publishers still generate as much revenue as they currently do on their own.

But Perplexity has certainly created a compelling product. Even if current evaluation has them looking set to lose the series of legal battles around AI overviews and fair use, the experience they’ve constructed for users seeking information and conducting research is unlikely to disappear. Writing this piece would have taken me a week if I had to go and find all that information listed above on my own, even with using Google search.

We’ll have to wait and see if the antagonization between Perplexity and news publishers results in any meaningful changes to their product or to the greater internet landscape.

With AI agents incoming that seem set to allow us to spend less time looking at our phones, the internet as a whole may be in for a paradigm shift away from this traffic-based subscription and ad revenue driven era.

❷

AI Agents — New Status Report

AI agents are emerging as a significant area of development and potential in the greater AI landscape. These autonomous systems, designed to perform tasks independently on behalf of users, are pushing the boundaries of what AI can accomplish. Recent advancements from major tech companies and startups alike are shaping the future of AI agents, promising both exciting possibilities and new challenges.

Anthropic's Claude: A Glimpse into the Future

Anthropic's recent demonstration of their AI agent, Claude, offers a compelling preview of what AI agents might be capable of in the near future. Claude can interact with a computer interface, taking screenshots, moving a virtual mouse, and typing to accomplish complex tasks. This level of autonomy represents a significant shift from traditional chatbot interactions, allowing users to delegate tasks rather than manage them directly.

Claude's capabilities were showcased through various tasks, including creating lesson plans and playing strategy games. The agent demonstrated an ability to develop and revise strategies, perform A/B testing, and operate for extended periods without human intervention. However, it also revealed current limitations, such as occasional errors in judgment and difficulty with nuanced decision-making.

Microsoft's AI Employees: Bringing Agents to the Workplace

Microsoft is taking a bold step into the world of AI agents with the introduction of their "AI employees" or virtual staff members. These autonomous agents are designed to handle a range of business tasks, from managing client queries to identifying sales leads.

The company is not only creating pre-built bots but also enabling clients to develop their own AI agents through the Copilot Studio platform.

This initiative aims to demonstrate the practical applications of AI in enhancing productivity and streamlining business processes. Early adopters, including McKinsey and Clifford Chance, are already exploring ways to integrate these AI agents into their workflows.

The Race for Advanced AI Agents

Other major players in the tech industry are also making significant strides in AI agent development.

Elon Musk's xAI has stated that it is working on advanced autonomous agents capable of performing complex tasks on behalf of humans.

As of this week, Google is also reportedly pursuing ‘computer-using agents’ with what they call Project Jarvis.

Challenges and Considerations

Despite the rapid progress, several challenges remain in the development and deployment of AI agents:

1. Reliability: As demonstrated by Claude's experience with the Paperclip Clicker game, AI agents can make errors that lead to inefficient or incorrect actions.

2. Ethical concerns: The development of autonomous AI systems raises questions about decision-making, accountability, and potential impacts on human employment.

3. User interaction: Finding the right balance between autonomy and user control remains a challenge, as does developing intuitive ways for users to guide and correct AI agents.

4. Scalability and cost: Current AI agents can be slow and resource-intensive, which may limit their practical applications in some scenarios.

The Road Ahead

As AI agents continue to evolve, we can expect to see more sophisticated systems capable of handling increasingly complex tasks. The focus will likely shift towards improving reliability, enhancing decision-making capabilities, and developing more intuitive user interfaces for agent interaction.

The integration of AI agents into various industries could lead to significant changes in how work is performed and how businesses operate. However, it's crucial to approach this technology with a balanced perspective, recognizing both its potential benefits and the need for responsible development and deployment.

AI agents represent a promising frontier in artificial intelligence, with the potential to revolutionize how we interact with technology and accomplish tasks. As research and development in this field accelerate, we stand on the brink of a new era in human-AI collaboration, one that will undoubtedly shape the future of work and daily life in profound ways.

❸

AI Edge Devices

“Edge devices” (like smartphones and other portable tech with small amounts of compute) are emerging as a crucial frontier for AI deployment.

Big Tech companies are racing to figure out:

1. How AI models can run locally on these smaller devices, rather than relying on cloud-based computing.

2. What consumers actually want.

Other than a few small failed projects like the Humane Pin and Rabbit’s R1, Google and Apple, as two of the major makers of the edge device most well suited to immediate AI integration (smartphones), are leading the way so far.

Apple's Approach: Cautious and Measured, or Just Lagging Behind?

Apple's strategy regarding AI has been a topic of debate in the tech industry.

While some view their approach as lagging behind competitors, others see it as a cautious and deliberate move. Apple has been focusing on on-device AI processing, prioritizing user privacy and data security. This approach aligns with their long-standing commitment to user privacy but has led to questions about whether they are actually just falling behind in the AI race.

With Apple Intelligence rolling out now, their Q4 results may provide some context for us in understanding mainstream consumer response to AI features. Will it be enough to save the ailing iPhone?

While Apple's AI efforts have primarily targeted the consumer market through features like Siri, Face ID, and computational photography, the company is also making inroads into the enterprise space. With the introduction of the M1 and M2 chips for Macs, Apple is positioning its AI-powered devices as productivity tools for businesses. The company's focus on privacy and security aligns well with enterprise needs, and its streamlined ecosystem can simplify IT management.

Qualcomm and Google: Revolutionizing Automotive AI

Qualcomm and Google recently announced a collaboration that aims to create advanced AI-powered digital cockpits for automobiles, leveraging Qualcomm's Snapdragon chips and Google's AI expertise.

Key aspects of this partnership include:

1. Development of a new AI framework for automotive manufacturers.

2. Integration with Android Automotive OS (AAOS).

3. Creation of custom voice-powered AI assistants for vehicles.

4. Introduction of new chips: Snapdragon Cockpit Elite and Snapdragon Ride Elite.

This initiative promises to deliver more intuitive and immersive in-car experiences, including advanced voice assistants, interactive maps, and real-time driving updates. The partnership also aims to optimize automotive software development for Snapdragon platforms, potentially reducing time-to-market for new features.

The Future of Apps in an AI-Driven World

Qualcomm CEO Cristiano Amon has presented an intriguing perspective on the future of applications in the era of AI. He suggests that the traditional concept of apps may evolve significantly as AI becomes more capable of understanding and executing user intent.

Amon envisions a future where users can simply express their needs to an AI assistant, which then composes the necessary "app" or interface on the fly. This could potentially decouple user intent from the constraints of traditional app structures, leading to more flexible and intuitive interactions with technology.

For example, instead of opening a banking app to check balances or pay bills, users might simply ask their AI assistant to perform these tasks, with the AI rendering the necessary information or executing the required actions seamlessly.

What’s Next?

As these technologies continue to develop, we can certainly expect to see more seamless integration of AI into our devices and daily routines.

However, questions remain about the long-term impact on traditional app structures, job markets, and user privacy. The coming years will likely bring both exciting advancements and new challenges as AI edge devices become increasingly prevalent in our lives.