143 | 🌴 ⚖️ FL & CA Bars Take AI Action

In today’s Brainyacts:

1. Bar action on AI ethics

2. What’s the deal with open source AI models?

3. M&A due diligence via AI and more AI model news

4. Government bureaucrats most threatened by AI and other related news

5. An AI palate cleanser

👋 to new subscribers!

To read previous editions, click here.

Lead Memo

Florida & California Bars take AI action

The Florida Bar has released proposed advisory opinion 24-1 regarding lawyers’ use of generative artificial intelligence. It hits on some of the key points and unique circumstances created by Generative AI such as it being similar to a non-lawyer human assistant. It is a short read and helpful. Here is a basic summary of what it covers:

1. Confidentiality: Lawyers must protect client confidentiality when using generative AI. This includes understanding the technology, especially if it's self-learning, to prevent inadvertent disclosure of client information. Lawyers should obtain informed consent from clients before using third-party AI that might disclose confidential information. They should also ensure that AI providers have strong confidentiality and security measures.

2. Competence and Oversight: Lawyers must understand the technology to satisfy their ethical obligations. They should review and verify the work produced by generative AI, ensuring it meets professional standards - just as they would with a human non-lawyer assistant. This includes being responsible for the accuracy and sufficiency of AI-generated work.

3. Ethical Delegation: Lawyers should be cautious about what tasks are delegated to generative AI. They must not delegate tasks that constitute the practice of law or require a lawyer’s personal judgment. When using AI for client intake, clear disclaimers and identification as AI are necessary to avoid misleading clients. NOTE: California is taking a more targeted approach to UPL (unauthorized practice of law).

4. Legal Fees and Costs: Lawyers must charge reasonable fees and costs, reflecting actual costs incurred. They should not duplicate charges or falsely inflate billable hours due to increased efficiency from AI use. Costs associated with generative AI should be transparent and justifiable. The language encourages the adoption of non-billable hour fee structures.

5. Advertising: Lawyers must adhere to advertising rules, ensuring that any use of AI in advertising is not misleading. They should clearly disclose when AI is used in client interactions and avoid making unverifiable claims about their AI’s superiority. Chatbots on websites are specifically called out here as they could easily be misinterpreted by the public.

6. Continuous Learning and Adaptation: As generative AI is a rapidly evolving field, lawyers should continuously update their knowledge and understanding of the technology, ensuring their practices align with evolving ethical and professional standards.

The California Bar is set to vote on AI Guidelines over disclosure and billing to determine if the unauthorized practice of law needs to be more clearly defined, and whether legal generative AI products require licensing or regulating.

The Committee on Professional Responsibility and Conduct’s (COPRAC) is calling for the board of trustees to work with California’s legislature and supreme court to determine if the unauthorized practice of law needs to be more clearly defined, and whether legal generative AI products require licensing or regulating.

Additionally, COPRAC released ethical guidelines (reminders mostly) on the use of generative AI but signals its continued evaluation and intention to make specific rule changes as the technology advances.

Spotlight

🏇🏎️ AI Models Get Better & Better: An open source primer

How do they measure how ‘good’ a model is?

There are many ways. One method is the MMLU, which stands for Measuring Massive Multitask Language Understanding. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more. See the complete list below. If you want to read the paper that explains this test, click here.

In open source models, what exactly is open?

Great question. There are many layers to an AI model: the underlying dataset, the model itself (how it has been trained/tuned), the parameters and weights (what use cases it has been designed to tackle), its interface and prompting context (how we interact with it). Each of these may be ‘open’ or not. Here is a basic diagram that shows how most ‘open source’ models are both closed and open.

There are generally two types of AI models: open source and private.

What is the difference between the two?

1. Performance Trends: As illustrated in the graph below, private AI models have historically outperformed open source models. However, this performance gap is gradually narrowing, indicating that open source models are catching up in terms of capabilities and efficiency.

2. Control and Customization: Open source AI models offer greater flexibility. Users have the ability to modify parameters and tailor the model to specific needs. In contrast, private models are controlled by the owning company, which sets the boundaries for what the model can and cannot do. While open source models may have fewer restrictions, any existing limitations can often be adjusted or bypassed by users with the necessary expertise.

3. Analogy with Power Systems:

   • Private AI Models: These are akin to using a traditional power utility service like Duke Energy. The service provider controls the production and quality of the power. In case of issues such as outages, you depend on the provider for resolution. Similarly, with private AI models, the owning company is responsible for maintenance, updates, and ensuring the model functions as intended.

   • Open Source AI Models: This is comparable to setting up an off-grid solar power system. While it offers more control and independence, it requires a comprehensive understanding and setup, including inverters, batteries, and other components. This independence comes with the responsibility of maintaining and understanding the entire system. Open source AI models offer similar autonomy but demand a higher level of expertise from the user for effective utilization and customization.

Making the Choice: The decision between open source and private AI models depends on various factors, including the desired level of control, specific use-case requirements, and the technical expertise available. With open source models rapidly improving, organizations now have a more viable choice between these two types of AI models, based on their unique needs and capabilities.

AI Model Notables

► LexisNexis Unveils Two New Generative AI Products

► AI meets M&A: How these former management consultants are upending due diligence

► Some of the UI features in Grok

► OpenAI is building next-generation AI GPT-5 — and CEO claims it could be superintelligent

► OpenAI DevDay Breakout Sessions are posted

►Google in talks to invest in AI startup Character.AI

► Getting emotional with ChatGPT could get you the best outputs

News You Can Use:

➭ Bill Gates: AI Is About to Completely Change How You Use Computers

➭ How Much Does It Cost to Train a Large Language Model? A Guide

➭ OpenAI's recruiters are targeting senior AI researchers at Google, offering annual compensation packages worth $5 million to $10 million

➭ AI threatens higher percentage of jobs in Washington, D.C., than anywhere else in US

➭ US-led AI declaration on responsible military use sees 45 countries join, but not China

➭ US and China set to forbid use of AI technology in autonomous weaponry

➭ France to host next AI safety summit as European nations jockey for tech leadership

➭ SAG-AFTRA Reveals How Studios Will Handle AI Replicas of Living and Dead Actors

➭ This AI robot chemist could make oxygen on Mars

Who is the author, Josh Kubicki?

Some of you know me. Others do not. Here is a short intro. I am a lawyer, entrepreneur, and teacher. I have transformed legal practices and built multi-million dollar businesses. Not a theorist, I am an applied researcher and former Chief Strategy Officer, recognized by Fast Company and Bloomberg Law for my unique work. Through this newsletter, I offer you pragmatic insights into leveraging AI to inform and improve your daily life in legal services.