Getting AI to critique previous AI work?

We make extensive use of AI and in the last 18 months or so that means OpenAI’s ChatGPT. We use it to guide both us and our clients’ activity in the digital space, but we never use it verbatim. It has a habit of just making stuff up, but being a Language Model it tends to create compelling output so it can be tempting just to take it at face value.

That’s a big no-no, and means we generally go back-and-forth checking stuff manually until we’re happy. We do, moybe ironically, see less of this hallucinatory stuff in custom-built GPTs but the output does need checking all the same.

Now, in a 180 degree spin, we decided to try and get an untrained GPT model to review content built by a trained model. Head gone yet? So, that’s a more hallucinatory AI model reviewing the work of a biased model (that is one aspect of what training essentially does). And off we go down the rabbit hole.

And, let’s start with something potentially contentious: our AI Policy from 2022. And this is what the general ChatGPT4 model produced when asked “summarise and critique” our page.

Connected’s AI policy and ethics page outlines their approach to leveraging AI in various aspects of their business, emphasizing labor-saving, insight, data-processing, economic growth, work nature, and innovation. They advocate for non-prejudicial systems, especially in computer vision and natural language processing, and adopt a “do no harm” stance. Their AI usage spans server management, predictive modeling, and marketing services for enhanced efficiency and accuracy. They highlight their openness to further AI advancements to improve client experiences. The content, created using AI, reflects a commitment to ethical AI use and continuous innovation.

For a detailed read, you can visit their website: [AI Policy & Ethics by Connected](

Critique: While the commitment to ethical AI use and innovation is clear, the document could benefit from more specific examples of ethical challenges and how they’re addressed. Detailing their methods for ensuring AI’s non-prejudicial application would enhance transparency. Additionally, mentioning external certifications or adherence to international guidelines could further solidify their stance on AI ethics.

I’m not sure if we have learned anything except that as time goes on, information changes and we should probably adopt a continuous improvement approach. Oh, and give more examples and show our working. Doh!

But, as we are here – why not use our current bespoke transformer to rewrite the page, or at least suggest how it should be written to make it more readable, compelling and in our “style”. Her is what the Connected AI trained model came up with, untouched and raw:

## AI Policy & Ethics: The Connected UK Manifesto

At Connected UK, we’re all about using AI responsibly, and we’ve got a plan to prove it. It’s not just any plan, though; it’s a manifesto. A declaration of our unwavering commitment to ensuring our AI doesn’t turn into the overlord of a dystopian future. We’re looking at you, Skynet.

### 1. The Human Touch

AI is smart, but let’s face it, it’s no match for the human brain (yet). We promise to keep humans in the loop, making the big decisions and ensuring our AI behaves. Think of it as teaching a very clever parrot to speak – someone’s got to make sure it doesn’t start reciting your credit card details.

### 2. Transparency: No Cloaks, No Daggers

Our approach to AI is like a glass house (but with better privacy settings). We’re all about being open on how our AI works, what data it uses, and why it’s convinced that cat pictures are the pinnacle of digital content. We’ll keep the jargon to a minimum and the explanations clear. Because the only thing our AI should be obfuscating is spam.

### 3. Bias? Not Here!

We’re committed to making our AI as unbiased as a coin flip. Our algorithms go through rigorous bias boot camp to ensure they treat all data fairly, whether it’s about your favorite ice cream flavor or your browsing history (we don’t judge).

### 4. Privacy: Our Fort Knox

Your data is your data. Period. We treat it like a state secret, ensuring it’s locked down tighter than Fort Knox. Our AI may be curious, but it knows better than to snoop where it’s not supposed to.

### 5. The Future: Bright but Grounded

We’re optimists at heart but realists in practice. While we’re excited about AI’s potential to revolutionize everything from your morning coffee to interstellar travel, we’re also committed to ethical guidelines that keep our ambitions in check. After all, we want to make sure we’re still around to enjoy the future we’re helping to create.

### Join Us on the Ethical AI Journey

At Connected UK, we’re on a mission to make AI as ethical as it is innovative. We’re looking for partners, clients, and fellow AI aficionados who share our vision for a future where technology enhances the human experience, without the side effects of rogue robots and questionable ethics. If that sounds like your cup of tea (or coffee, we’re not picky), let’s connect!

Honestly, I’m not convinced as to the style for what is a serious subject (more training required, I suspect) and it wasn’t to provide any more useful depth. I suspect, content such as this will have to stay in human hands … at least until GPT5xx or whatever comes next.