In today’s Brainyacts:

1. Renaming OpenAI’s “custom instructions”

2. Bing Chat solicits competitive intel in chat sessions

3. Judges get free access to my course

4. Recent developments of leading AI model updates

5. News you can use (links to click if the headline grabs you)

6. Yoda!

👋 to new subscribers!

To read previous editions, click here.

Lead Story

👩‍🎨⚙️ It’s not you, it’s them: The Paradox of Progress: Generative AI's Continuing Design Dilemma

Poor Design Again

In the ever-evolving landscape of technology, generative AI models stand out as one of the most promising and transformative innovations. However, as with any nascent technology, its rollout and adoption are fraught with challenges. One of the most glaring issues is the design choices these models make when introducing new features and capabilities. Instead of simplifying and democratizing access, they often end up creating a labyrinth of technoAI jargon that leaves the average user bewildered.

Historically, poor design choices in technology have manifested in various ways, from Microsoft’s annoying and useless “Clippy” to Google Wave. Generative AI models are no exception. While they promise unparalleled capabilities, their interface and communication often fall short of being user-centric.

Take, for instance, OpenAI’s term, 'custom instructions.' On the surface, it sounds like a feature that offers customization—a buzzword in today's consumer-driven market. However, dig a little deeper, and one realizes that it's not as straightforward as it seems. Custom instructions, as introduced by OpenAI for ChatGPT, allow users to provide specific details for their interactions with the chatbot. It's a feature that promises personalized and relevant conversations.

But here's the catch: the term itself is shrouded in ambiguity and confusion. What does "custom instructions" mean to a layperson? Without context, it sounds technical and intimidating.

🤡 The irony is palpable. A feature designed to enhance user experience is presented in a manner that many might find off-putting. It's akin to offering a delicious meal but serving it on a plate that's too hot to touch. The essence of the feature gets lost in translation, overshadowed by its complex nomenclature.

Today's users are savvy. They're accustomed to intuitive designs and interfaces that require minimal learning curves (thank you Apple and your iPhone). They expect their AI tools to be as reliable as a seasoned professional from the get-go. But with great power comes great responsibility. As these models become more advanced, with features like custom instructions, there's an inherent risk of miscommunication and misuse - something we definitely do not need more.

It's a double-edged sword: while these tools become more potent, they also become more complex.

The solution? A paradigm shift in design philosophy. Generative AI models need to prioritize user-centric design. This means doing away with technoAI jargon and adopting consumer-friendly language. It means providing clear, concise instructions and ensuring that features like "custom instructions" are presented in a manner that resonates with the average user.

Why not ask OpenAI’s GPT-4 for some help?

▶︎▶︎PROMPT

What might be some better terms to use instead of custom instructions?

▼▼Response

Certainly! The term "custom instructions" can be ambiguous for the average user. Here are some alternative terms that might be more intuitive and user-friendly:

1. Personalized Settings: Emphasizes the personal touch the feature offers.

2. Chat Preferences: Directly indicates that it's about setting preferences for the chat.

3. Conversation Guide: Suggests that users can guide or steer the conversation in a direction they prefer.

4. Interaction Blueprint: Implies a plan or framework for how the chat will proceed.

5. Chat Tailoring: Indicates customization of the chat experience.

6. User Profile Setup: Suggests that users can set up a profile with specific details for the chatbot to remember.

7. Chatbot Memory: A simple term indicating what you want the chatbot to "remember" about you.

8. Conversation Customizer: A direct indication of customizing the conversation.

9. Chat Blueprint: Similar to "Interaction Blueprint," but more concise.

10. My Chat Rules: A more informal term that gives users a sense of control over the conversation.

It's essential to test these terms with a diverse group of users to determine which one resonates the most and is the most intuitive. The key is to find a term that is self-explanatory, easy to remember, and gives users a clear idea of the feature's purpose.

🐾🐶🍲

I mean how hard was that OpenAI? Perhaps you should eat your own dog food and use your own tool just to start with? 😆👍🏻

For good measure, I asked Bing Chat a similar question. It replied with:
I agree that the name could be more clear and intuitive. Maybe something like “chat preferences” or “chat profile” would be better.

Come on, this is too easy!

Spotlight Story

Bing Chat's Curious Quest for Knowledge

When crafting the essay above (Lead Story), I turned to Bing Chat since OpenAI isn't currently web-connected (though it briefly used Bing for a few months). My exploration centered on OpenAI's "custom instructions" feature.

Interestingly, when I posed a question to Bing Chat, it responded with a few queries of its own. This led to an enlightening exchange, revealing its intent behind those questions.

Here's a snippet of our conversation:

We continued a bit more . . .

It seems Microsoft is keenly observing Bing Chat's interactions, particularly when rival models are mentioned. However, considering Microsoft's significant investment in OpenAI and its use of a GPT-4 variant for Bing, it's more of a collaboration than competition. This piqued my interest.

Yet, it's also possible that this was just another typical AI-generated conversation, not a deliberate design choice in Bing Chat. Maybe it's just one of countless fleeting AI chats that fade into the digital ether.

The world of generative AI is truly captivating. After all, in this realm, what can we truly claim to know?

Free Access to My Course for Any Sitting Judge

In light of invaluable discussions with state ethics committees and inspired by insights from Ralph Artigliere, a revered retired judge from within the Brainyacts community, I am excited to present a special offer: free access to the "Generative AI Blueprint" course for every sitting judge, not just in the U.S. but globally.

If you are an active judge or are acquainted with one, I encourage you to reach out. Simply send an email to [email protected], confirming your current judicial position. This is a decent opportunity to get up to speed and understand the power (and risks) of conversational generative AI in the legal realm.

AI Model Notables

• ChatGPT expands its ‘custom instructions’ feature to free users

• Anthropic launches improved version of its entry-level LLM

• GPT-4’s reasoning ability questioned – these researchers say that it can’t

• Paxton.ai releases free access to all US federal and state laws

News You Can Use:

➭ How this Canadian injury lawyer uses AI

➭ White House offers prize money for hacker-thwarting AI

➭ US Federal Elections Commission (FEC) taking up AI in political ads to thwart deepfakes

➭ Speaking of deepfakes, this is crazy 👇

➭ Let Consumers Take AI to Court Empower the People to Police AI

➭ Open letter from news outlets demanding rules for AI training data

➭ Poll shows Americans distrust AI

➭ The Pope speaks on AI