We all recognize that the value and benefit of AI comes from the data (LLMs and Training Models), however, the present challenge is the access to internal, historical, project documentation (e.g. charters, risk/issues logs, project schedules, lessons learned, etc.). And, if a PM/PMO has access to this information there is the concern of privacy associated with it e.g. don't use it outside the organization. What is your present access to this internal data? Saving Changes...
If an organization implements ChatGPT Enterprise or similar products from other providers where there is a commitment made to protect their data and where fine-tuned models are built solely for use by the organization and not part of the publicly-available ones, then the risk would be at the same level as with any other cloud-based solution provider.
At that point, it is really a question of the national (i.e. government), industry-level and company-specific risk appetites. If those are extremely conservative, then at best, sanitized, redacted data could be used.
Access is usually not the concern for PMO staff in private sector organizations as relevant data tends to be housed in shared information repositories with not too granular access controls. The bigger concern is around compliance with the organization's data policies.
Kiron
...
1 reply by Larry Gagnon
Jan 03, 2024 5:09 PM
Larry Gagnon
...
Thanks for the response. Agree with your good points. I think the bottom line are the "risk appetites" of organizations. A point I was attempting to raise, is the value of combining internal with external models (obviously, I didn't do a good job of that when posting my question).
If an organization implements ChatGPT Enterprise or similar products from other providers where there is a commitment made to protect their data and where fine-tuned models are built solely for use by the organization and not part of the publicly-available ones, then the risk would be at the same level as with any other cloud-based solution provider.
At that point, it is really a question of the national (i.e. government), industry-level and company-specific risk appetites. If those are extremely conservative, then at best, sanitized, redacted data could be used.
Access is usually not the concern for PMO staff in private sector organizations as relevant data tends to be housed in shared information repositories with not too granular access controls. The bigger concern is around compliance with the organization's data policies.
Kiron
Thanks for the response. Agree with your good points. I think the bottom line are the "risk appetites" of organizations. A point I was attempting to raise, is the value of combining internal with external models (obviously, I didn't do a good job of that when posting my question). Saving Changes...
Currently, if it is confidential or proprietary, our AI has no access to that information which includes process documentation. Essentially, any server, database, web page, document, etc. that is access controlled beyond our external firewall is off-limits.
I'm sure our AI specialists have data for experiments, but I am sure it is sanitized and tightly compartmentalized. Saving Changes...
Sergio Luis ConteHelping to create solutions for everyone| Worldwide based OrganizationsBuenos Aires, Argentina
The same you stated here. Projects are started to create solutions to put strategy in the field. Because on that, in my previous and actual experience, just some solutions or some components of the solutions was able to be published. For example, if publishing them gave value to the organization. Saving Changes...
Good to hear about organizations' efforts to make data available (or not). I do think that the research frontier is finding the blend between internal and external data sources. Saving Changes...
kotresh taralePM Consultant| Eximius Design Bangalore, Karnataka, India
In my experience I see that use of chatGPT for new data can allow it to augment information from both internal and external sources. I mean what it already learned can't be taken away and it will only adds upto what the information we are going to feed. So risk analysis must be done when we are feeding specific information to the tool each time. New or Custom LLM or LLM modelling tools could become solution in this case. Saving Changes...
