It's a common fact that ALL LLM models were trained on a public accessed repositories, which are licensed either on GPL or MIT/BSD/Apache licenses. Unless AI companies are explicitly breaking the law they can't train their models on proprietary code.Thus, it's only fair to force all LLM code providers to license generated code under GPL/AGPL since it's impossible to differentiate if data for generated code was licensed on GPL or BSD.
>>108637623AI models should be open source.
>>108637670They are already open-source since they're trained on open-source data. You're already have all access to all data it was trained on.What you meant is probably they're must publicly available like Deepseek? Well, that would be great, but idk what you're going to do with 600B model, it's no like you can self-host it.
>>108637688>idk what you're going to do with 600B model, it's no like you can self-host it.I will simply wait until it finishes processing a response.
spoiler: they are 100% training on proprietary codethe data sets are not sanitized based on licensing or anything like thatif the code is source available, it goes in the modelso yes, the government should 100% seize these companies and make them publicly owned, because they stole everyone’s shit, and should not be allowed to profit from it
>>108637688If the weights were available, it could be trained & also audited for backdoors/biases/censorship.Obviously that would cost a lot of money but that's besides the point.
>>108637792>spoiler: they are 100% training on proprietary codeThey only train on proprietary code that is available to them. Most proprietary code resides in private self-hosted repositories on corporate servers that you can't from WAN.
>>108637623AI companies don't need to obey laws.
>>108637623>outsourcing your brain is...LE GOOD
