Is your feature request related to a problem?
Tensor4all has a very well thought-out plan (see email below by Hiroshi) about how they deal with copyright, attribution, and agents. We should discuss this, then adapt something similar.
Hi all,
The discussion at yesterday's developer meeting was very useful.
Clarifying copyright and scientific credit is important for our community, so I have now systematized it in two ways.
(1) Agent rules.
The shared rule set that our AI agents load automatically (tensor4all-agent-rules) now includes explicit rules on copyright,
on recording reference sources in the code, and on respecting scientific credit:
https://github.com/tensor4all/tensor4all-agent-rules/blob/main/rules/common/provenance.md
In short: code written while referencing a third-party project must keep a reference to the source in a comment;
close translation keeps the original copyright and license; and when scientific credit is due beyond the license,
the agent proposes an entry in the citation policy after confirming with the user.
For a concrete example of what this looks like in practice, see the derivation notice at the top of our TDVP planning module, which was written closely following ITensorNetworks.jl and therefore carries an Apache-2.0 notice:
https://github.com/tensor4all/tensor4all-rs/blob/main/crates/tensor4all-treetn/src/tdvp/plan.rs
These rules are loaded not only by the agents each of us uses, but also by the AI bot that audits the tenferro-rs codebase.
So compliance is checked continuously, not only at implementation time.
(2) Recursive citation policy.
tensor4all-rs, strided-rs, and tenferro-rs now each have a "Provenance and Citation Policy" document:
https://github.com/tensor4all/tensor4all-rs/blob/main/docs/PROVENANCE_AND_CITATION_POLICY.md
https://github.com/tensor4all/strided-rs/blob/main/docs/PROVENANCE_AND_CITATION_POLICY.md
https://github.com/tensor4all/tenferro-rs/blob/main/docs/PROVENANCE_AND_CITATION_POLICY.md
Each document records which components build on which upstream projects and in what way,
lists the original papers of the implemented algorithms, and asks research users to cite those papers and to apply the citation policies of upstream projects recursively.
This makes the contributions from upstream projects visible in each repository, and encourages proper scientific credit to flow to them.
None of this is enforceable, but I believe it is the best we can do at the moment.
Comments are welcome, and corrections to the provenance tables are especially appreciated
Best wishes,
Hiroshi Shinaoka
Describe the solution you'd like
discussion of the points in the email
Describe alternatives you've considered
n/a
Additional context
We can also ask Hiroshi directly.
Is your feature request related to a problem?
Tensor4all has a very well thought-out plan (see email below by Hiroshi) about how they deal with copyright, attribution, and agents. We should discuss this, then adapt something similar.
Hi all,
The discussion at yesterday's developer meeting was very useful.
Clarifying copyright and scientific credit is important for our community, so I have now systematized it in two ways.
(1) Agent rules.
The shared rule set that our AI agents load automatically (tensor4all-agent-rules) now includes explicit rules on copyright,
on recording reference sources in the code, and on respecting scientific credit:
https://github.com/tensor4all/tensor4all-agent-rules/blob/main/rules/common/provenance.md
In short: code written while referencing a third-party project must keep a reference to the source in a comment;
close translation keeps the original copyright and license; and when scientific credit is due beyond the license,
the agent proposes an entry in the citation policy after confirming with the user.
For a concrete example of what this looks like in practice, see the derivation notice at the top of our TDVP planning module, which was written closely following ITensorNetworks.jl and therefore carries an Apache-2.0 notice:
https://github.com/tensor4all/tensor4all-rs/blob/main/crates/tensor4all-treetn/src/tdvp/plan.rs
These rules are loaded not only by the agents each of us uses, but also by the AI bot that audits the tenferro-rs codebase.
So compliance is checked continuously, not only at implementation time.
(2) Recursive citation policy.
tensor4all-rs, strided-rs, and tenferro-rs now each have a "Provenance and Citation Policy" document:
https://github.com/tensor4all/tensor4all-rs/blob/main/docs/PROVENANCE_AND_CITATION_POLICY.md
https://github.com/tensor4all/strided-rs/blob/main/docs/PROVENANCE_AND_CITATION_POLICY.md
https://github.com/tensor4all/tenferro-rs/blob/main/docs/PROVENANCE_AND_CITATION_POLICY.md
Each document records which components build on which upstream projects and in what way,
lists the original papers of the implemented algorithms, and asks research users to cite those papers and to apply the citation policies of upstream projects recursively.
This makes the contributions from upstream projects visible in each repository, and encourages proper scientific credit to flow to them.
None of this is enforceable, but I believe it is the best we can do at the moment.
Comments are welcome, and corrections to the provenance tables are especially appreciated
Best wishes,
Hiroshi Shinaoka
Describe the solution you'd like
discussion of the points in the email
Describe alternatives you've considered
n/a
Additional context
We can also ask Hiroshi directly.