Openness Pragmatism

Since open weight is at least better than no weights, people are worried about excessive criticism of models that do not meet the standards of those who are more rigorous with the definition of open source. To further this point, some value weights more than the training data. Creating a model from scratch—while resource-limited to do so—is not actually important for this community (re: widder_2023).

• Open weight is better than nothing. (better_than_nothing, 8) ^betterthannothing8
  • Somewhat related to ^opensourceweight13. It isn’t necessarily saying that open weight is open source, but that it is better than closed source. Also related somewhat to ^modelsnotoss1
• Weights are more valuable than training data, unlike traditional OSS where the source is more important than the binary. Highly related to ^modelsnotoss1 and ^modelsaresoftware1.(weights_more_valuable, 6)
• Arguments over what is and isn’t open is pedantic and unnecessary. Related to ^morepedantic (argument_open_pedantic, 6)
• Excessive criticism of limited openness coming from models built by companies like Meta is only going to stifle development of even open weight models. (excessive_criticism_bad, 3)
• Restrictive licenses are still permissive towards the vast majority, as the commercial threshold that such licenses set are much greater than the economic value an individual would produce through the use of an open model. This calls into question whether simply classifying through licenses (as shown in ^openthroughlicense7 is practically useful. (restrictive_license_permissive, 2)
• The work of impressive proprietary model builders is as laudable as the individual’s AI OSS project. The point is that those publishing open model tools on GitHub are as deserving of praise as those at OpenAI building ChatGPT. (work_equally_laudable, 2)
• Weights are not code, therefore should not be thought of as the same thing. ^modelsnotoss1 is the equivalent in Terminological Rigor. (weight_not_code, 1)

In some ways, these comments may be suggesting that people want more ownership over what openness means for them, instead of prescribed answers.

• It is good to be critical about groups claiming ownership to a definition of open source, and be open minded about its application to open models. Stands in opposition to ^opentrustedsources3. (critical_ownership_open, 2)
• Aspects like dataset, code, and documentation has never been a requirement for open source. (documentation_open_source, 1)