• bss03@infosec.pub
        link
        fedilink
        English
        arrow-up
        1
        ·
        42 minutes ago

        Yeah, pretty much. When LLMs (and GPT architecture) was just “in the lab”, and were honest-to-goodness research, even if done almost entirely within non-governmental organizations, scraping the whole Internet for training was defensible (tho, still not entirely consensual) as fair use.

        But, the moment the commercialized it in any way, they should have been slapped with a copyright-based injunction against distributing any model, any model output, and against training any further models. The oroborus of investment that inflated the markets never would have happened.

        Failing that once the cyclic investments were revealed, dozens of fraud and conspiracy to commit fraud cases should be been opened.

        If somehow you accept their unreasonable copyright use and unreasonable accounting, they still have civil and possibly criminal liability for the text generated by their products when used directly. When their products are licensed and used by other companies and deployed as mandatory parts of sale and support flows, then either they or the company deploying them (or both) as also liable for the output of those products.

        I think underlying the technologies might very well be useful and ethical in the future, but the industry as it exists today is criminal and corrupt from the human rights abuses doing training and throughout.