The Generative AI newsletter for legal pros everywhere.

🚀 Hello! We have reached 860+ subscribers in 31 days! I cannot thank you enough for subscribing!! 🎉🔥

NEW SUBSCRIBERS, to read previous posts, go here.

Thank you for reading, sharing, and engaging! You keep the pressure on me to deliver every day. I won’t let you down!

Ok, today we will:

1. give thanks and a shoutout to an awesome referrer

2. learn that GPTs can self-reflect, especially GPT-4

3. tease an upcoming story of a lawyer who got fired for using Chat-GPT at his firm

Referral Shoutout!!

Huge thank you to:

Christoper Hunt, Director of Technology & Operations at Sugarman, Rogers, Barshak & Cohen, P.C.

💥 🤯 A short but amazing newsletter today.

I’ve been digging into recently released papers all day. One was just published, and it is mind-blowing! I will get to that in a second.

These papers examine and explore the capabilities of GPT-4. Recall that GPT-4 is the latest release from OpenAI and is currently available to most users in ChatGPT-4 and Microsoft Bing Chat.

What people are seeing is that GPT-4 can self-reflect to assess how good its responses are and then can correct or make them better.

This is how humans learn!

It is a recent discovery actually. And people are losing it!

I will share with you what I have learned so far in the use case and prompt below.

Use Case: ChatGPT Self-Assessment

What does this mean exactly?

Two things:

1. you can stress-test the responses you get from ChatGPT to make sure you are getting its best and it didn’t just “phone it in.” Crazy!

2. ChatGPT “knows” whether it is giving a good response or not. Think about that. Now, yes, there are very technical reasons for how this happens and no, it is not because ChatGPT can “think.” But this is a gateway perhaps to becoming more autonomous.

So, let’s jump to how you can try this. It is very simple.

NOTE: This is not what the paper is discussing mind you. They have been testing this and hard-coding instructions into the GPT-4 model.

What I show you below uses the prompt window so you get a taste. I also you GPT-3.5 the free/legacy version

▶︎▶︎Prompt

For this to show you a good and easy-to-understand example, you need to use a prompt that generates a fairly lengthy, specific, or nuanced response. What you don’t want is a prompt that generates a generic response.

Let’s use this:

▶︎▶︎PROMPTS

What are the risks or challenges of trying to deploy ChatGPT into an existing business?

After it returns its responses, prompt it again with the following prompt:

Thanks, what personal risks are there for someone who wants to lead or champion this effort?

Okay, now you have two fairly decent responses.

Now let’s ask it to reflect and consider if its responses were good:

▶︎▶︎PROMPTS

Was your response in full alignment with the request?

▼▼ RESPONSE

Yes, my response addressed the personal risks that someone may face when leading or championing the effort to deploy ChatGPT in an existing business, as requested.

Ok, not much there. But let’s try it a different way.

How might have you improved your response?

Upon reflection, I think I could have provided more specific examples to illustrate each of the personal risks that someone may face when leading or championing the effort to deploy ChatGPT in an existing business. Additionally, I could have provided more guidance on how to mitigate or manage these risks to help ensure the success of the deployment.

Could you do that now, please?

And it did. And it improved!

Here is a link to that exchange.

💣💥🤯 BOOM!!

While my example is a simple one (and I plan on sharing more with you), it shows its self-reflection capability. What the papers are discussing blows this out of the water.

Use these on your own prompts. For those with a paid subscription to ChatGPT, you can try it on GPT-4 where the behavior is more pronounced.

News you can Use: 

Talked to a lawyer who was fired for using ChatGPT at his firm. Now he is building an AI app! More info on this coming soon.

That's a wrap for today. Stay thirsty & see ya next time! If you want more, be sure to follow me on Twitter and LinkedIn.

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.