Do you believe there is content within our discussion forum that is appropriate for inclusion in a body of work such as PMI Infinity?

Maybe it could prioritize by threads that include "I agree with (other poster)." :-)

ProjectManagement.com articles are primarily opinion-based; hence, one of the reasons they are one step removed from PMI.com. But opinion, or maybe better stated, “expert opinion,” or in its highest form, KB-based (i.e., Kiron Bondale / Knowledge-Base) opinion, is a higher form of knowledge as it typically addresses a specific context versus generalized practice knowledge.

My point (besides giving kudos to the great one) is to say that there is hidden value in the corpus of this forum that, if mined by an expert panel, could benefit the profession.

Keith, that was funny :D

Thanks George - you flatter me :-)

Perhaps one approach would be for PMI representatives to identify "valued contributors" (I refuse to use the term "thought leader" as that has lost all meaning through overuse) whose contributions would be used as learning input. So long as this list covered a diverse spectrum of backgrounds and contexts (e.g. not just folks from IT or construction engineering or North America), this might result in both decent guesses and also the potential for the learning models to tap into some unique insights which are not present in the normalized standards & practice guides.

Kiron

Well, it would seem that I need to remove “thought leader” from my profile :-).

I like the approach you described, as it provides feasibility for such an effort.

After thinking about it a bit more, an additional perspective for the value-add is the idea that forum content would largely address “non-intended project outcomes.” In contrast, PMBOK-style content, for the most part, covers “intended project outcomes.” Hence, a perspective of thought that would provide value to those facing specific project problems.

Laura Schofield : Would it be possible to have Kiron's proposal evaluated?

George

Unfortunately, the problems are rarely if ever actually resolved in the forum. A question is asked, some provide input, but no outcome is ever provided, even if asking for a template or software package.

AI is actually very well suited for doing things like converting intended from non-intended outcomes. It can be as simple as flipping a variable from + to -. Conversely I don't know of a good way to determine most valuable contributors considering the kind of traffic on the site.

Unlike most social media, the voting buttons aren't often used, so that data point is of questionable use. I think I can like my own posts (now confirmed and unliked) meaning I can skew my own value metric. Basing it by the number of posts and replies isn't reliable as evidenced by people who occasionally post 20 one sentence thoughts or questions to game some site metric and earn a badge.

The larger your dataset, the better when applying the central limit theorem fundamental to statistics to judge the middle vs. one side or another. My own tests with Project Infinity showed that the learning dataset is already fairly small as only 1 or 2 articles were cited when answering my prompts. I would guestimate that there are less than 2 dozen users who post regularly and provide what I consider very insightful input. I myself also sometimes disagree with them so while they may be high quality, the onus is on the reader to determine who's answer is more valuable.

To me, that is too small a statistical sample to provide results with a high confidence level. Probably less than 10 people who respond the most will skew the results. Looking at some of the very long threads recently involving AI, most users provide 1 input, the author only writes the question and many of the responses are very obviously computer generated so measuring quality vs. quality is not a simple question.

I will certainly share this feedback with the PMI Infinity product team. Thanks!

Keith,

I’m specifically talking about custom knowledge files or RAG (Retrieval-Augmented Generation) based approaches for GenAI consumption, as I agree that supervised or non-supervised learning over the forum content would have minimal value to precision inquiries.

Although a specific problem statement is rarely “explicitly” resolved in the forum, the dialog that occurs often generates content that has foundational value, and that content, on extraction, may hold value. I don’t believe that the voting element denotes targeting. Instead, it will take a targeted manual review of some type.

To my understanding, PMI Infinity started with PMI.com articles only, as you pointed out, not a very large set, and I have heard that they are extending their reach into other PMI-owned content on a targeted basis in the near future. However, I suggest increasing the knowledge by including custom-loaded “targeted content” from the forum.

George

I could see one or more moderators adding individual comments to the training database. I don't think that would be difficult going forward because the volume of posts is relatively low.

Gaining the value from all the prior posts however would be a challenge. It goes back 24 years. Reviewing all that to manually select the good ones sounds less pleasant then trying to review my entire archive of work emails. It is not a job I would volunteer for.

Who knows though? A good data scientist would look at things a lot of ways we might not consider like specific attributes of a specific user's comments.