A ChatGPT/Bard Warning Story

Or why using LLMs to do research into anything you do not have sufficient knowledge of is dangerous

After all the breathless reporting about how LLMs are going to remake business as we know it, I decided to use both ChatGPT & Bard (you know, Google’s ChatGPT) to help me research towards a new venture I’m working to launch, a democracy technology venture generator.?

To give you background, the crisis of democracy Israel has struggled with over the past five months has convinced me, among others, that representative democracy has been hacked by special interests and illiberal ideologies. If we want to live in a world free from coercion, we need to safeguard and strengthen liberal democracy and that requires upgrading our democratic practices.

Moving from Democracy 1.0 to Democracy 2.0 we will need to expand the ease of adoption of democratic approaches such as participation and deliberation through tech enabled services.?

Given my experience over the past twenty years building social venture accelerators ( PresenTense ), a VC-backed mission driven technology company ( MobileODT ), and a commercialization platform for public-purpose technologies ( CoVelocity ), I’ve come to believe that we won’t be able to generate quality ventures aligned with these public goals without aligning purpose and profit.?

Hence the idea to build a venture generator focused on launching technology-enabled services to help transition our public sector into participatory democracy.?

With that background in mind (which I hope you can abstract to your work in terms of why I can’t afford to get the basic building blocks of the venture right) my warning story:?

Yesterday, after reading victory stories about how ChatGPT can summarize financial statements, how it can process full-scale dissertations into a few take home bullet points, how it can pass the Bar and Medical License Exams, I thought to myself: hey, why are you conducting a competitive analysis by yourself?

Let Chatty take care of it!

So I did. I wrote a prompt and sent ChatGPT (and Bard, because it’s ‘net connected) off to the races. Here is the prompt: “Please build me a table of for-profit startups or companies who are building technology-enabled services to strengthen democracy. Please exclude companies who are nonprofit or not-for-profit, or companies where you cannot find information on their investors. Please focus on startups. Include in the table a description of the product, a description of the business model, and a list of their top three investors.”

And the results? They looked spectacular!!

I noticed differences between Bard and ChatGPT and even submitted a snarky ticket to Bard saying, “hey, ChatGPT still eats your lunch.” And I cut and pasted the table into my working doc, and decided I could use the extra hour in my day to eat a bowl of cereal.?

As I was eating my cereal, I reflected on the world and our role in it. I thought about how we would all soon just be the arms and legs of the superorganism directed by an LLM to do the work that it identifies as most mission critical based on its super sophisticated generative modeling.

Which got me thinking: perhaps I didn’t even need to start the venture generator because, well, ChatGPT could surely just write and launch those companies on its own if the AI decides that we need Democracy 2.0.?

I finished my cereal - didn’t need the whole hour for a bowl, it turns out - and decided to kill time by googling the companies it listed, to see how they’re doing.?

And it turned out that ….they didn’t exactly exist.?

Sorry, that’s not the right word - some of them did exist, at some point, until they closed, and some of them are actually awesome in their own right. But the information ChatGPT and Bard presented with supreme confidence in an official looking table was ….wrong. Just wrong.?

Thank goodness I ate that bowl of cereal quickly enough to give myself time to check, and I didn’t shoot off a document to my partners bragging, “look how fast I can do the job I said would take me all day.”

The point of all of this is to say: LLMs in their current state are great toys. Fun to play with. Good for qualitative creative sketching, for perhaps copy editing your text, for inspiring the application of wisdom to the world. They may even help us complete bullshit jobs where truth or accuracy or veracity are not the goal.?

More than that though? Depends how much you depend on the truth.?

**Just for fun I popped this human-written post into ChatGPT & Bard and asked them to edit it so it would be catchy for LinkedIn. The results …kinda scary. Both played up the value of LLMs, inserting marketing language to make them more enticing. Adding what Chatty would have had me write you to give you a taste: