Proposal for automated checks of knowledge submission

[katesoule]

In order to release knowledge contributions to open source community, IBM legal is asking us to propose an automated solution for our generation pipeline that covers:

On the provided knowledge source from the contributor:

• HAP detection
• PII detection
• Screening for commonly copyrighted content (e.g. OneShield) -- lower priority

On the synthetic data created from the knowledge source

• Screening for leakage of text from the knowledge source being used verbatim in the synthetic data (leveraging OneShield)

[lehors]

What's HAP? I proposed a PR template for the taxonomy repo (see taxonomy PR 27) and was considering adding a question along the lines of:

Can you confirm that your contribution is original, not based on data from another system, and not encumbered by any copyright or other IP restrictions?

But I'm sure a lawyer would be able to get the right words in. Obviously this isn't bulletproof but it might worth adding?

[katesoule]

HAP is Hate, Abuse and Profanity. We will want this actually tested when people are submitting long documents that can't be fully read by the reviewers, and not just have people check a box.

[katesoule]

Legal is coming up with the right wording, it will most likely be added through a CLA.

[lehors]

HAP is Hate, Abuse and Profanity. We will want this actually tested when people are submitting long documents that can't be fully read by the reviewers, and not just have people check a box.

Thanks. Speaking of which, shouldn't we set some scope for what kind of contributions would be welcome? For one thing, I assume, for now at least, we are limiting contributions to English language, right? I think we should make that clear in the documentation so that people are aware.

[lehors]

Legal is coming up with the right wording, it will most likely be added through a CLA.

CLAs are a huge deterrent to contributions...

[darrellreimer]

@katesoule I had been thinking we would be doing these checks on the server side, not in the cli so this would be transparent to users of lmdk. We (the ones building the models and taking knowledge in) would run hap/pii/copyright etc, and provide a reference implementation to show how and what we're doing.

The fact that IBM legal is recommending we do this isn't really visibly to anyone - it's just a reasonable practice for us

[xukai92]

second to @darrellreimer seems too much for the LMDK/CLI side

[hickeyma]

@katesoule Following the feedback from @darrellreimer and @xukai92, can this issue be closed as not applicable to CLI?

[tuhinsharma121]

yeah. second to @darrellreimer

[anik120]

Putting this on hold since it's unclear if there's anything that needs to be done in the cli. My initial read is "no".

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had activity within 60 days. It will be automatically closed if no further activity occurs within 31 days.

[juliadenham]

Closing as we now work with Red Hat legal.