ChatGPT Use Policy

[Ismael-KG]

Please leave any thoughts on Kairoi's Template ChatGPT Use Policy below. For example:

• How can the introduction be made clearer?
• Should the template make more explicit reference to concerns arising from inappropriate use of ChatGPT? And how?
• Should the template be more or less specific than it currently is, and why?

You may read the latest version of the template here: https://github.com/KairoiAI/Resources/blob/main/Template-ChatGPT-policy.md.

There are also some slides providing context to this template here, as well as reasons for having such a use policy on our blog.

[HuwWDay]

I really like this! Just a few suggestions: 1) wondering if you wanted to account for other LLMs and how they might be used differently. For example ChatGPT shouldn't be used for research because it just makes up citations, but Bing Chat could be used for this. Bing Chat and Bard are the main two ones I'd consider writing about. 2) You've written what is expressly forbidden and what is acceptable but might it be worth writing what is sorta inadvisable and it might even be worth suggesting why/at the very least if you are going to do these things that aren't forbidden, how should you do them in the best way possible (e.g. make sure you can verify a translation/written code some way) 3) The main two uses I think you should mention in more detail is translation (akin to google translate) and writing code.

All in all I think it's really good!

[Ismael-KG]

Amazing! Thank you @HDiscoDay! Point by point (splitting point 1 in two):

1.1. Yes! If I had capacity, I would make all the LLM Use Policies 😅 But a better use of people's time is a more inclusive policy. Hence, this one-liner (which may need improving – what do you think?):

Whilst this policy is targeted at ChatGPT, you are encouraged to apply it to other AI systems that may be in use at your organisation.

1.2. I just shared a summary mentioning the need for I might need for the template policy to be adapted to organisation's contexts and relevant practices (e.g.: research can mean very different things). I think this general guideline covers this but please let me know if it needs clarifying (or another guideline altogether):

ChatGPT can provide inaccurate information, and all its responses must be keenly checked for errors

2. I try to cover "sorta inadvisable" with the general guidelines, but this might be worth clarifying in the intro of the template. And examples (check translations and code) should be added by users of the template. I think there's a lot to do for the intro/framing of the template 😅

3. See 2. Translation is interesting but I think there are better proven tools for that. Maybe an extra guideline is "ask yourself and others whether there are better tools for the task at hand". And coding is a very particular skill and job. The template can mention these use cases but I would not assume all organisations have coding going on in-house.

[HansF]

Hey, thanks for this, I've done a few pulls wrt some formatting.

A friend told me it's a great way to "Get past the big white sheet."

• Ask for the outline of reports if it is the first report of the sort (for example: “What's a useful outline of a recruitment report?”)
• Ask for the outline of a blogpost for inspiration (for example: I need to write a blogpost introducing privacy-enhancing technologies – how could it be structured?”)

Could become:

• it's a great way to 'get past the big white sheet'. Ask for guideance or inspiration when trying something new (for example “What's a useful outline of a recruitment report?” or I "need to write a blogpost introducing privacy-enhancing technologies – how could it be structured?”).

[HansF]

I want to translate it for my co-corkers. Thank you for the permissive license,

In order to facilitate the process, may I kindly request that you include a basic Creative Commons attribution line to the document that I can easily copy, including your preferred name, email, and/or URL. This would be greatly appreciated.

[Ismael-KG]

@HansF that would be marvellous! I have this in the README but have chnged it slightly for the template file. Do you think this reads okay?

Based on Template ChatGPT Use Policy by Kairoi Ltd (2023) | kairoi.uk | CC-BY 4.0

Once I have more feedback, I will make a PDF or Word file with clear acknowledgements!

[Ismael-KG]

Just to note that @nickbarroww suggests a paragraph right after the first "Context and Scope" one explaining how using systems like ChatGPT entails risks; they are not infallible but can make things up ("hallucinate" if you use that term) be biased, and use the data you provide1 for further training (there is a privacy/security issue), etc.

Does anyone have any thoughts on what this paragraph might look like?

[HansF]

I've had a go at it:

LLM's are trained to produce 'credible' answers, one of the ways they learn is by using the feedback from the user on the responses they give them. It is possible that when you are using chatgpt you are training it. One of the characteristics of LLM's is that they can be completely wrong, whilst 'sounding' very confident in their output. So always double-check!

Thinking on metaphores that might connect with larger audience, maybe you can use them:

• Bit like the gps app on your car, you trust it but at the same time you're responsible for driving
• a smart but inexperienced intern, that does prep work but that needs to be checked before it's executed/sent out.

[Ismael-KG]

Thanks @HansF, I'll get back to this shortly. I just wanted to bring UNESCO's new report to this project's attention, as it outlines a really nice list of uses of ChatGPT in higher education (page 9). I don't think these are entirely pertinent to the average workplace, but can be helpful to keep in mind

[letykemp]

Hey, this is great, thank you for sharing.

I also agree that coding assistance should be in the Acceptable Uses heading, so I'd suggest further developing it to reflect how useful ChatGPT can be for formulas, syntax usage, command explanation, error clarification and step-by-step instructions on any software.

You mentioned that organisations would not have in-house coding but this kind of assistance are also valid for non-programming software such as Excel, Google Sheets, Slack, Teams etc.

Hope this helps, it's looking really good!

[GemmaTuron]

Hi @Ismael-KG !

this is a great and valuable resource for many organisations, including us at Ersilia. To start with, I have created a fork of your repo and modified the template to adapt it to our basic needs (currently, just shortened it and changed the coding example for something more relevant to us) - I'll share with the rest of the org and see if everyone is happy with this initial version, we will then iterate on it.

Perhaps you could ask people who adopt this now, to share their use cases so we can all learn what works for different orgs, if people is happy to share I'm sure we can learn from the community. Here is ours: https://github.com/GemmaTuron/Resources/blob/main/Template-ChatGPT-policy.md

[Ismael-KG]

Ah, this is wonderful, thank you @GemmaTuron! Thank you so much! I actually asked someone earlier today for just this! I'd love to write up a blogpost about the different adaptations 😊

[ghost]

It is possible to be translated in an Asian language like Filipino? This would benefit a lot of organisations in the Philippines. What's the process like for translation

[Ismael-KG]

Hi @bohemianchic! Thank you so much for asking! I would love for you to translate this to other languages!

There is no formal process. As the resource is made available under CC-BY, anybody can make a translated version and simply acknowledge and respect the template's license by adding the following line in a clearly visible part of the new document:	"Based on Template ChatGPT Use Policy by Kairoi Ltd (2023) / kairoi.uk / CC-BY 4.0"

It would be great if you could translate it. I wouldn't mind if you fork the repo and make the template your own (with attribution), or if you want to raise a pull request here with a translated version (so it might be referenced as "Kairoi Ltd. (2023) translated by Jean Dinco"). It really is up to you! Also happy to arrange a call about this 😊

[Ismael-KG]

Thanks @letykemp for chatting just now – the quick thought is to acknowledge that, as ChatGPT evolves, past conversations will become part of the data it draws on for your interactions. This might be a good argument for employer-specific accounts that keep your personal usage separate.