ChatPTFD: Publish the fine documentation!

Hi fellow humans and excellent people,

as you might have heard ChatGPT, GPT-4 and LLMs make quiet some noise now (e.g. Bengio et al. 2023, Gebru et al. 2023, Politico), which involves the tech scene but also a growing number of professionals from other disciplines and laymen. This is not only about energy consumption, reliability, transparency, data protection and ethical issues of application (see e.g. Bender et al. 2021), but also about intellectual property and the origins of creativity.

As this field might also touch physical models and engineering designs in the near future (see Point-E, Runway Gen-2, DreamFusion), this might be a chance to advocate for open-source solutions in the field - let’s say in another open letter or onepager.

One could argue for

  • making open-source works the basis of those model’s training data and publish products under equal licenses to respect and simultaneously overcome IP issues.
  • promoting open-source solutions in software, hardware, (physical) models, image and language products to increase availability of appropriate and harmless training data.
  • establishing and expanding infrastructure for “global commons in science, technology and innovation” (see Lynn & Salzman 2023) to leverage technology success in a broad, participatory and egalitarian way.
  • creating human-machine systems and collective intelligence, that enables sustainable digital fabrication and knowledge transfer.

Is this a topic for GOSH? What are your thoughts and additions? Would love to hear/read that!