Discussion Thread

I've actually been hearing more and more buzz about this topic, so I want to revisit this thread.  Lately, I find ChatGPT to be a powerful motivator for me to be worthy of my own neurons.  There was an entertaining and informative episode on the podcast "Cautionary Tales" about chatbots that got me thinking about the role of AI in our industry (link below).  The takeaway is that it's much easier to pretend to be human than to actually do the things that we value in humans.  And a bot that fakes it well probably isn't representing the best traits of humans in the first place.  Remember Tay?

You can ask ChatGPT any simple question with a technical dimension and it will give an answer that probably sounds convincing to a client and will produce totally ineffective results if implemented.  OpenAI recognizes this in the "Limitations" section of their "Introducing ChatGPT" post (link below).  I've had a lot of conversations lately where I've pointed out that ChatGPT and other text-generating bots aren't intended to be correct or honest about anything, and heard in response some version of, "well it's just not quite there yet."  As if this is even a step towards that.  In fact, ChatGPT has not even started becoming accurate.  That's not what it's for.

Do engineers-and people in general-talk a lot without saying anything?  Rely on careful wording to hide our lack of expertise on a subject?  Parrot information without review or synthesis?  Certainly, and engineered systems have failed as a result.  If we're excited that ChatGPT sounds convincingly human, maybe we should also be a little disappointed that humans have so much in common with a piece of software that is not programmed to tell the truth, provide any needed service, or generate original informed opinions.  Maybe the question isn't, "can AI do my job?", it's, "what should I be doing that makes me better than AI?"

Links:
Tim Harford, Cautionary Tales: "The Online Date That's Too Good to Be True"
https://timharford.com/2022/10/cautionary-tales-the-online-date-thats-too-good-to-be-true/

OpenAI: "Introducing ChatGPT"
https://openai.com/blog/chatgpt

I've actually been hearing more and more buzz about this topic, so I want to revisit this thread.  Lately, I find ChatGPT to be a powerful motivator for me to be worthy of my own neurons.  There was an entertaining and informative episode on the podcast "Cautionary Tales" about chatbots that got me thinking about the role of AI in our industry (link below).  The takeaway is that it's much easier to pretend to be human than to actually do the things that we value in humans.  And a bot that fakes it well probably isn't representing the best traits of humans in the first place.  Remember Tay?

You can ask ChatGPT any simple question with a technical dimension and it will give an answer that probably sounds convincing to a client and will produce totally ineffective results if implemented.  OpenAI recognizes this in the "Limitations" section of their "Introducing ChatGPT" post (link below).  I've had a lot of conversations lately where I've pointed out that ChatGPT and other text-generating bots aren't intended to be correct or honest about anything, and heard in response some version of, "well it's just not quite there yet."  As if this is even a step towards that.  In fact, ChatGPT has not even started becoming accurate.  That's not what it's for.

Do engineers-and people in general-talk a lot without saying anything?  Rely on careful wording to hide our lack of expertise on a subject?  Parrot information without review or synthesis?  Certainly, and engineered systems have failed as a result.  If we're excited that ChatGPT sounds convincingly human, maybe we should also be a little disappointed that humans have so much in common with a piece of software that is not programmed to tell the truth, provide any needed service, or generate original informed opinions.  Maybe the question isn't, "can AI do my job?", it's, "what should I be doing that makes me better than AI?"

Links:
Tim Harford, Cautionary Tales: "The Online Date That's Too Good to Be True"
https://timharford.com/2022/10/cautionary-tales-the-online-date-thats-too-good-to-be-true/

OpenAI: "Introducing ChatGPT"
https://openai.com/blog/chatgpt

I've used Chat GPT for a variety of applications. The best use case I've found is for generating social media captions. I know most larger firms can afford a marketing team, but if you're a smaller consulting firm, you may do your own social media in-house. I know I've always had trouble writing interesting captions or explaining things in a way that the average person can understand. Fortunately, ChatGPT and other AI chat programs can provide succinct descriptions and narratives using proper grammar. 

I've also used Chat GPT to help plan presentations. By asking Chat GPT to give you a "mind map" on a particular topic, it can help you organize your thoughts for your presentation in a way that people will understand. It can even be asked to provide some questions relating to your presentation topic to help you plan ahead for what an audience might ask. 

I've also heard of people using Chat GPT to role-play a job interview in order to give them practice answering questions. 

Chat GPT has a lot of use cases, but it's always important to read the response thoroughly and provide edits where needed. The tool isn't perfect, but with a little review and editing, you can end up with a really great product.