One of my unsettling concerns that I have not yet heard mentioned is... who controls the models?
At least with one major LLM model, when it started outputting political messaging inconsistent with the views of the owner of the model, they made sure to "fix" it to align with their own worldview... Large and powerful entities control most of the more powerful LLM models available, and those entities have the power to adjust the output to their own ends... Probably not on a case-by-case basis, sure, but I could see any number of things potentially detrimental to the Free and Open Source Software commons being intentionally injected into these models... to meet the interests of the owner, not the person at a keyboard typing at a prompt. Even with a paid subscription, I get the impression that the true cost of operating these servics is nowhere near what these for-profit entities are charging... they appear to be operating at a loss in order to build a base... and they are almost certainly not doing that out of a humanitarian interest, given that they are large for-profit entities... and in many cases, we do not have to infer, they have spelled out their intentions, however outlandish (e.g. thinking as a utility, anyone?) For a truely open model with independently curated training sets, perhaps this would not apply, but by and large I suspect that is a minority concern in this discussion... When I swing a hammer, I expect that the limits of my hand-eye coordination, prior practice and skill be the factors determining how successful I am at driving a nail... not the manufacturer of the hammer who happens to be able to decide which nails I am allowed to hit successfully, let alone if I hit my own thumb. live well, vagrant
signature.asc
Description: PGP signature
