Post B5DbvtTpcvXbL4L4gC by divVerent@misskey.de
 (DIR) More posts by divVerent@misskey.de
 (DIR) Post #B5DbvtTpcvXbL4L4gC by divVerent@misskey.de
       0 likes, 0 repeats
       
       @Suiseiseki@freesoftwareextremist.com @SuperDicq@minidisc.tokyo Actually not quite. You can get the software to train your own models, and it can be open source. Open sourcing such a model however would require providing all the training data...
       
 (DIR) Post #B5Dbvtdl21Trpqz09Q by djsumdog@djsumdog.com
       0 likes, 0 repeats
       
       I don't think there are any "open source" models. OSI has granted the label open source to a bunch of models that are just open weight or "free as in beer." Yes, for a model to truly be open source, you need all the training data and the tools anyone can run to regenerate their own version of the model. But most of the training data is either closed/proprietary or falls under one of 2-bajallion open source licenses. The AI trainers simply do not care.