090 | 📻♟️ 10 Queries to Tune Your AI Strategy

In today’s Brainyacts we:

1. will experiment with a Brainyacts AMA (ask me anything)

2. share big numbers on the Generative AI market

3. offer 10 query use cases to help with your AI strategy

4. apply undetectable watermarks to AI content

5. ask who to sue with AI-generated music fan tributes

👋 A special Welcome! to NEW SUBSCRIBERS.
To reach previous posts, go here.

📩 🎁Your Invitation

As I near the end of my 100-day commitment to writing this daily newsletter, I'm feeling more driven than ever to delve deeper into the pragmatic use of generative AI in the legal ecosystem. The last 90 days have been an immense learning journey for me, and as I look forward, I realize that I don't have to undertake this journey alone.

So, let's experiment together. I want to invite you to participate in an experimental event - the Brainyacts AMA (ask me anything). This will be a first-of-its-kind, salon-style Mastermind group, hosted online and free of charge. To maintain the intimacy and quality of discussion, participation will be limited to 8-10 select individuals.

The goal is to foster dynamic, collective discussions around the practical implementation and future implications of generative AI technology. I believe this is a unique opportunity for us to share insights, learn from each other, and co-create the future of our industry.

If you're excited about engaging in these critical discussions and ready to join this thought-provoking experiment, I encourage you to express your interest. I have no idea if I'll receive one reply or a hundred. If the response is high, I may follow up with a few questions to aid selection.

Please know this isn't about social sorting or excluding anyone – it's about creating the most effective discussion group possible. Your kindness and patience are appreciated as I navigate this process.

To express your interest, reply directly to this email or reach out to me at [email protected] Please share why you want to join, what perspectives or experiences you bring to the table, and a question or topic you're eager to explore.

Let's see if we can achieve critical mass.

💰📐 The Numbers

Two sets of numbers for you to consider today.

1. The first is the potential market size of all things generative AI.

Bloomberg Intelligence expects the generative AI market to soar and fuel a decade-long boom, with the market for generative AI reaching $1.3 trillion in 2032 from $40 billion last year.

To give you some sense of size, the total global automotive industry consisting of passenger and commercial vehicles, is somewhere between $2 to 3 trillion USD. And this market has been in development for over 100 years!

2. The second set of numbers comes courtesy of Cathy Wood of ARK Investment Management (for more on her click here). She provides insight into the falling costs of developing LLMs.

Cathy recently provided an in-depth analysis of the generative AI market. Click here (video) and here (report) for resources. In both she shares this intriguing finding:

If OpenAI were to have develop ChatGPT in 2015, it would have costs roughly $800M.

In 2022 when it did develop it, the cost was roughly $5M.

If OpenAI were developing ChatGPT today, the cost would be less than $300k.

🤯 And if they were to develop it in 2030, it would cost about $30!! This is about a 70% decline in cost YOY.

This is a staggering amount of cost reduction and means only one thing - that the development, numbers, and sophistication of LLMs is going to exponentially increase.

Takeaway: hold on tight as we are only in the very beginning of where this all will go.

🧐🤖 Use These Queries to Shape and Stress Test Your Generative AI Approach and Starting Point

As I've engaged with numerous individuals across the legal sector and beyond while writing this newsletter, I've gained valuable insights into utilizing LLMs (Legal Language Models) in organizations. In today's issue, I'll discuss the various types of queries or use cases your team should contemplate, prioritize, and develop scenarios for, to understand their potential applications and significance.

A systematic approach to these queries will help you refine your generative AI strategies and identify stronger starting points. Simply adopting a broad-based horizontal generative AI strategy may result in disappointments and false starts. A more targeted, tactical starting point is a smarter approach. Leveraging these queries will enable you to tailor your use cases based on their relevance to your operations.

I've listed 10 unique types of queries, each with example goals or potential questions that could arise in using them.

As you review these, reflect on your current methods of answering these questions. Consider the resources involved - the effort required, the personnel responsible, and the data needed. Generative AI, with its ability to generate answers instantaneously through prompts, could be the game-changer you need.

1. Temporal Reasoning: This allows the firm to understand past trends and forecast future scenarios.

   • Example 1: "What were the revenue trends for our firm over the past five years?"

   • Example 2: "Predict how our service demand might change in the next fiscal year based on current market trends."

2. Summary: This can help synthesize lengthy reports or crucial information for quick understanding.

   • Example 1: "Can you summarize the key findings of this market research report?"

   • Example 2: "What are the main points in this new client’s Outside Counsel Guidelines?"

3. Fact-based Question Lookup: This can aid in recalling or searching for specific business facts.

   • Example 1: "What was our Q2 profit margin last year?"

   • Example 2: "Who are the current top three competitors in our practice area?"

4. Compare and Contrast: This could be useful for comparing different business strategies or evaluating competitors.

   • Example 1: "What are the differences in services offered by our firm and Firm X?"

   • Example 2: "Compare the soundness and clarity of this legal argument with that of our opposing counsel’s.”

5. Text to SQL: This can assist in handling databases and extracting valuable business insights.

   • Example 1: "Find all clients from California who have had fees for services under $100,000 in the last fiscal year."

   • Example 2: "What's the average client retention rate in the labor and employment practice over the last 5 years?”

6. Hypothesis Generation: This can be useful for brainstorming solutions to business problems or exploring potential growth opportunities.

   • Example 1: "Why might we be seeing a decline in repeat clients in our NYC office within the financial services sector?"”

   • Example 2: "What could be some new markets for our consulting services related to legal operations advising?"

7. Simulation of Dialogue: This can be used for practicing negotiations, client meetings, or managing difficult conversations.

   • Example 1: "Simulate a conversation with a client who is unsatisfied with our services."

   • Example 2: "Imagine a negotiation for a new engagement with a potential big client who demands fixed fees for substantial portions of work."”

8. Creative Writing: This could be used for drafting engaging marketing copy or crafting company-wide communications.

   • Example 1: "Compose a compelling LinkedIn post announcing our new consulting services."

   • Example 2: "Write a motivating email to the team ahead of a trying and intense trial prep and war room effort."

9. Translation: This can assist in handling international clients or interpreting documents in foreign languages.

   • Example 1: "Translate this contract from Spanish to English."

   • Example 2: "Translate our service brochure into Arabic."

10. Emotional Understanding: This can help respond to client communications or feedback more effectively.

    • Example 1: "A client has written a dissatisfied email about our service..." (The LLM could suggest appropriate response strategies.)

    • Example 2: "We've received positive feedback from a big client!" (The LLM could help craft a response to foster further engagement.)

These examples illustrate how LLMs can offer significant value in professional service contexts. They can help analyze data, streamline communications, foster creative problem solving, and much more.

🚨Why are these important? Because not all LLMs can tackle these queries with the same might. Both inputs (prompts, training, injections, etc.) and outputs will be different. Knowing which queries are your most vital will help you ask better questions and decide on a more suitable LLM approach.

News you can Use: 

Watermark Generative AI Content

In the fast-paced world of AI, generating text that looks like it was written by a human is becoming increasingly easy. But how do we tell if a piece of text was created by an AI model or a human? Traditional methods are becoming ineffective, especially as these models evolve and produce near-human-quality text.

This paper proposes an intriguing solution: embedding 'watermarks' into AI-generated text, much like the ones used in currency and artwork. However, it's essential that these watermarks do not degrade the quality of the text or become too obvious. Hence, the aim is to create a watermark that remains undetectable and maintains the same quality as the original model.

Moreover, this watermark is detectable only with a secret key, which ensures that even if a large amount of random information is used to create a response, it can still be recognized as coming from the watermarked model. This approach underlines the importance of maintaining the integrity of the original model without falsely accusing humans of using AI to generate their texts - an issue which has already raised controversies.

Finally, the paper presents the formal definition and construction of undetectable watermarks, and introduces a concept it calls 'empirical entropy' to measure the randomness used in generating specific outputs.

 Here is the paper.

Question: If I Sing in the Style and Actual Voice (Using AI) of a Deceased Singer, But Do So as a Tribute and Not to Sell, Who Can Sue Me?

Sounds like a bar exam hypothetical. But it is real. This is one of the questions being raised as AI generated music fans are resurrecting singers, like John Lennon, who have long since passed, to show their appreciation and share it with other fans.

This article discusses the intriguing yet complex legal issues arising from the use of artificial intelligence (AI) in music creation, using the example of AI-generated Beatles songs. The key legal concerns revolve around potential copyright infringement, the right to publicity, and the application of 'fair use' doctrine. The article points out the necessity for clearer industry standards or regulatory measures, possibly driven by voluntary standards or litigation, to navigate these grey areas. Additionally, the potential backlash from suing over fan-created tributes that aren't meant to be monetized is also highlighted.

DISCLAIMER: None of this is legal advice. This newsletter is strictly educational and is not legal advice or a solicitation to buy or sell any assets or to make any legal decisions. Please /be careful and do your own research.8