Revolutionizing Jira Administration: The Impact of ChatGPT

[Kyle Rector]

ChatGPT has the potential to significantly impact Jira Admins by automating repetitive tasks and providing quick and accurate answers to common questions. This can save time and increase efficiency for Jira Admins, allowing them to focus on more important tasks. For example, ChatGPT can be integrated into Jira to answer questions about project status, assign tasks to team members, and provide updates on ticket progress. Additionally, ChatGPT can be used to automate tasks such as creating new tickets, updating ticket status, and logging time spent on tasks.

One major advantage of ChatGPT is its ability to understand natural language, which makes it easy for Jira Admins to communicate with the system. This means that Jira Admins can simply type in a question or task in plain English, rather than having to use specific commands or navigate through menus.

Another benefit of ChatGPT is its ability to learn and adapt over time. As Jira Admins use the system, ChatGPT can learn their preferences and workflows, becoming more efficient and accurate in the tasks it performs.

Overall, ChatGPT has the potential to revolutionize the way Jira Admins work by automating repetitive tasks, providing quick and accurate answers, and adapting to individual workflows.

Please be noted that Jira is currently not integrated with ChatGPT, but you can use GPT-3 based integration solution to integrate with Jira.

Thoughts?

[Nic Brough -Adaptavist-]

The AI detection routines flagged this posting as having a 99.4% chance of being fake, and one of the things I'm hoping to get added to the next version of community is a "detect chatGPT and not allow people to even post it".

Very simply put, it's useless at the moment, and please stop using it to generate your posts.

You are not the first person to use it to post here, but I would say that you're the first to get it to generate a half-way decent post (but not answer).  The last one I'm aware of generated around 80 answers before we stopped them, and every single one of them was, at best, useless. 

It lies, it invents functions, it mis-reads the documentation, and it really has no clue that people are talking about different systems.

[Stephen Coleman]

I'm sorry, but you're mistaken. Discussing how an upcoming technology could improve an existing one shouldn't be discouraged.

Is it flawed?
Yes, but so are coworkers and forum participants.

Does it confidently deliver wrong answers?
Yes, but so do people.

Its value will be based on additional development and how it is used, which should absolutely be discussed in an open forum, especially one that's focus-driven among the very sort of people that would have an interest in how it is used.

Instead of shunning the new technology, you can be a logical voice of opposition. Explain to people who have an interest in the technology the pitfalls they could encounter, and that will lead to higher learning and possible solutions to those problems.

[Nic Brough -Adaptavist-]

I am sorry, I should have emphasised the "at the moment"

I'm not shunning it at all, I'm just pointing out that it's nowhere near as clever (and hence useful) as a lot of its proponents are saying it is.

>One major advantage of ChatGPT is its ability to understand natural language,

That's a huge fallacy.  It does not "understand" natural language, it is simply able to respond to it.  

It's very clear that it does not understand it, because it turns out answers that, although quite well written, are utter garbage.

>This means that Jira Admins can simply type in a question or task in plain English, rather than having to use specific commands or navigate through menus.

Again, no they can't.  To get even a vaguely useful answer out of it, you have to be incredibly specific and use the jargon for the subject you are talking about.  For the more vague and simple questions (i.e. the vast majority of ones people are likely to ask), it's no better than hitting a search engine with keywords.

I totally agree that there could be huge potential here.  But it's a long way off being useful in the present - it needs years of training with negative feedback - people who do understand the subject telling it where it gets it wrong and correcting it.

[Stephen Coleman]

You're not wrong. That's exactly what it needs. But the discussions of how it can or can't be helpful and the warnings of the pitfalls to others shouldn't wait until it gets there.

>It's very clear that it does not understand it, because it turns out answers that, although quite well written, are utter garbage.
So do some executives.🤣

Maybe rather than shutting down the conversation, it can be funneled to a proper home to keep it going.

[Nic Brough -Adaptavist-]

@Stephen Coleman  I'm totally with you on all of that.

We should be talking about it

• A lot of humans do indeed turn out utter garbage answers too
• Yes, this is a conversation we should not shut down.  I don't think I emphasised "it could be useful" enough in my posts.  I really don't want to stop talking about AI (until it starts getting the basic answers right)
• The negative bit - no, we can't start using bad AI to answer questions.  ChatGPT hasn't been trained on Atlassian stuff, so it's a bad AI to use (nor have any of the others)
• If it turns out bad answers, it needs to be stopped, like we do with the humans who get it wrong.
• It's not a case of funnelling answers, it's a case of stopping bad answers.

I'm going to be brutally honest - I ask ChatGPT to answer questions on non-Atlassian fora.  Not directly, it gives me an answer and I edit it to make it more "me".  But it's useless on Atlassian stuff.

[Paulo]

@Nic Brough -Adaptavist-  I wouldn't say that it's completely useless. 

Heres a video of ChatGPT potentially saving 20min of my time writing Jira REST API calls: https://www.linkedin.com/posts/pauloaaramalho_chatgpt-jira-jiraservicemanagement-activity-7034653336909791232-DTvH

But I mostly agree with you, and I'm not expecting it to answer and solve Jira Service Management tickets yet, at least not until OpenAI or someone else decides to launch a service that can learn from Confluence articles and propose answers, and I honestly wouldn't trust it. However, seems like some people are already playing with it :p See also this post on the group Jira Community on LinkedIn : https://www.linkedin.com/feed/update/urn:li:activity:7034544918769364993

[Nic Brough -Adaptavist-]

I think you've answered your own question there.  It potentially saved you 20 minutes of coding.  But you had to spend hours checking that it had got it right, and in the vast majority of cases, it does not.

It needs a lot more training on every subject people might point it at.  I am not saying it can't help us in the future.  I get 500 emails a day about Atlassian stuff and at least 350 of them could easily be answered by either a well-trained expert system (or AI) or by LMGTFY.  I'd love an AI that can do that.

But we've tried it, and it fails, and fails very very hard.

It's not that it can't help, but we are actively looking at ways to detect ChatGPT answers so that they can be marked wrong and removed already.  It really is that bad.

[Pablo Beltran _Marketplace Expert_]

My 5 cents.

I understand some people are afraid of their jobs and they are aggressive against such amazing technology.

AI works. And works today, not tomorrow or within a couple of years.

However, it requires a lot of effort and money to fine-tunning a LLM model to solve a specific problem.

AI is a huge business. And public ChatGPT is a showcase for marketing and attract potential customers to OpenAI.

Soon, you realize that if you want to use ChatGPT (or Google Bart/Palm2) for the real World,  you need to spend a lot of money fine-tuning it :) 

@Kyle Rectoryou are right. And that future is coming quickly.

Please be noted that Jira is currently not integrated with ChatGPT, but you can use GPT-3 based integration solution to integrate with Jira

Atlassian has build "Intelligence" application that works pretty well. Did you try it? And a lot of Marketplace partners have integrated ChatGPT with Jira

I was very skeptical about AI some few months ago until we built our first AI-powered app. I can say now based on real facts, that today AI works, and works quite well. 

[Nic Brough -Adaptavist-]

So @Pablo Beltran _Marketplace Expert_ how do you explain that ChatGPT has only answered one Atlassian software question usefully of the 1,200 we have asked of it?

[Pablo Beltran _Marketplace Expert_]

@Nic Brough -Adaptavist- : Did you make questions against an "out-of-the-box" foundation model? Or against a fine-tuned one?

What kind of questions do you mean? JQL?

We used a fine-tuned ChatGTP 3.5 turbo model. 

In our particular case, we refused to use the ChatGPT knowledge of JQL. Therefore, I cannot speak about how much precise is ChatGPT in that field. However, we trained ChatGPT to learn JQL from the scratch, and surprisingly this approach worked very well.

I mean if you are disappointed with default answers from foundation models (JQL for instance), you can "overwrite" such part by fine-tunning the model, and take advantage of the power of LLM to understand human natural language only. Then use translation capabilities for what you need:

User's prompt in NL -> LLM -> (what you need)

I encourage you to try our Ask Jira app, indeed. This is not for marketing purposes. IMHO, You will become less skeptical if you see AI working against Jira Software.

There is also a native ChatGPT plugin named JiraTalk claiming to do something similar, but I did not try it deeply yet.

[Nic Brough -Adaptavist-]

Everything, not just JQL.

[Pablo Beltran _Marketplace Expert_]

Understood.

The gap is:

Can you train (customize) LLM to get precise answers? YES, and organizations have to pay for that.

Can foundational LLM produce precise answers "out-of-the-box"? a LOTTERY.

The OpenAI marketing has worked as many people think that ChatGPT works out-of-the-box.

However, you must invest quite a lot to make ChatGPT work as your expect.

how do you explain that ChatGPT has only answered one Atlassian software question usefully of the 1,200 we have asked of it?

You made a great step forward as you already have 1200 useful prompts (questions).

Now, you need to write the expected response for each one (completion) and create a JSONL file to fine-tune ChatGPT.

Of course, things are not so much simple, but this is the first step.

May I ask you share those prompts? We are very interested in them!

We can fine-tune ChatGPT to cover all of them.

At the end, you will see how really AI works and it is capable to answer ALL OF THEM with amazing precision.

We are willing to collaborate. Hopefully you are interested too.

[Nic Brough -Adaptavist-]

Try getting it to answer questions on the Community (do not post the answers, the moderators already have enough to do correcting or removing the terrible answers 'bots have been posting)

You're going to need to train it on a lot more than 1200 good answers, and bear in mind that you're going to have to continue to train it as the software changes.

[Pablo Beltran _Marketplace Expert_]

To fine tune chatGPT you may want to read this OopenAI documentation:

https://platform.openai.com/docs/guides/fine-tuning

Happy training!