ChatGPT and AI: will it transform peer-to-peer communities like ours?

[ChrisHemedinger]

Hello SAS Community!

Like many people, I'm fascinated (and concerned) with the AI advancements in large language models like ChatGPT. I spent time learning about how it works and how effective it can be in answering questions about SAS and other topics. I wrote an article about it: ChatGPT brings AI into popular culture.

I feel the value of this community is not only the problem/question->solution interactions, but in the real-world experience that informs those questions and answers. Community members get to know each other, in a way, and build trust in the expertise offered by known contributors.

StackOverflow -- a programming-focused forum that some people also use for SAS questions -- has taken the step of "banning" GPT-generated replies. I haven't seen this to be a concern on this community, but we're watching!

I have no doubt that AI-based expert systems will have a big impact on our personal and professional lives. The key is to be intentional about what we allow it to do for us (to save us time and effort), and reserve for ourselves the high-value interactions that help us to learn and grow.

I welcome your thoughts -- reply here or leave a comment on my blog article. Are you excited about the potential of ChatGPT? What are your concerns?

[Quentin]

I'm interested/excited/amused by ChatGPT.  

I wouldn't worry about banning it from here, unless there is an obvious influx of humorously incorrect answers.  

I swear I saw a post somewhere a couple weeks ago (LinkedIn)? Where they said to ChatGPT "You are a SAS terminal window" or something like that.  And then they submitted stuff like:

proc freq data=sashelp.class;
  tables name;
run;

And ChatGpt actually made some fairly close predictions for the appropriate response.  I wish I could find the link.

That said, as a non-AI guy myself, I think you make the key point in your blog post: ChatGPT is about prediction, not thinking/logic. But as with other machine learning stuff, when prediction is good enough, it can be very useful.  Even without an explainable logic model.

[PaigeMiller]

A recent example where ChatGPT gave a sub-optimal reply to a very simple question, and also had incorrect syntax.

https://communities.sas.com/t5/SAS-Programming/Extracting-year-and-month-from-date9-date/td-p/856421

[Quentin]

---

@Quentin wrote:

I swear I saw a post somewhere a couple weeks ago (LinkedIn)? Where they said to ChatGPT "You are a SAS terminal window" or something like that.

I found the post:

https://www.linkedin.com/pulse/make-chatgpt-think-its-live-sas-session-aka-code-submits-herrmann

Wild stuff.  Just tested it, and it works.  If anyone wants to try it, here's the starting text I used (copied from the article):

[DEBUG_MODE: ON; You are an AI that acts as a Linux (Ubuntu) system terminal. All responses from here on out are terminal responses to my input in your responses, please escape the $ sign, so that it doesn't render strikes. Do not add any further commentary or explanation. Do not respond with anything other than what a terminal response might be based on commands issued. The "sasdemo" useer has just logged in.]

[MarkusWeick]

One aspect:

If I think about google translator and DeepL, I see two useful (low level) AI tools. I use them quite often, but still check and regularly adjust the results.

Interpolating this to ChatGPT: might be an useful instrument but with limitations.

[MarkusWeick]

Another aspect (and I think the more important one) in our context.

With traditional information/communication we have got something like a 2-factor-model: The information itself and our image of the source of the information, as for example the cultural background or the personality of the source.

With ChatGPT I am still missing the second part.

[ballardw]

I wonder what ChatGPT would say about the quality of the code for ChatGPT...

[MarkusWeick]

Just ask it 🙂