3 ms·

It’s frustrating too seeing some people in AI say the technology has no negative implications. Of course it does. Foreign actors are using it to influence elect

It’s frustrating too seeing some people in AI say the technology has no negative implications. Of course it does. Foreign actors are using it to influence elections. But when Jeremy Howard was asked if he had done impact assessment of a large open source model release, he replied (roughly) “No. Impact is about the same as a new release of pencil and paper.” Which… actually that’s not true. Iran isn’t using pencil and paper to influence US elections, they’re using LLMs. And if all the big cloud providers cut them off, they’ll move to using home brew models. That is the impact and putting our head in the sand about it isn’t going to be helpful. This is NOT to say we should not release to models! I think we should release open models, and that in addition to that we should also care about written impact assessments so everyone is on the same page about risks! I think it’s great that Jeremy Howard is releasing open foundation models, but I wish he wasn’t so flippant about the real world implications. I’ve always looked up to him for the work he has done, and found that attitude very disappointing.

CatWChainsaw 2y ago
Also frustrating to be constantly patronized that the negative consequences will definitely be outweighed by positive ones, pinky promise. Right now we have threats of job insecurity in basically every industry, Noah's Arc levels of spam, the undermining of democracy as a functioning form of government, and insane energy demands and environmental consequences. But on the plus side you can outsource the entire design team to one guy tweaking SD outputs and you can use an LLM to summarize the spam generated by a different LLM, so all's well that ends well I guess.
- TaylorAlexander 2y ago
  Agreed. I think the short term profit motive is really blinding people to the risks. Like we could have continued this in a research setting for longer to really get a sense for how to deal with these things, but folks wanted to productize because of profit potential backed by incredible hype.

CatWChainsaw 2y ago

Also frustrating to be constantly patronized that the negative consequences will definitely be outweighed by positive ones, pinky promise. Right now we have threats of job insecurity in basically every industry, Noah's Arc levels of spam, the undermining of democracy as a functioning form of government, and insane energy demands and environmental consequences. But on the plus side you can outsource the entire design team to one guy tweaking SD outputs and you can use an LLM to summarize the spam generated by a different LLM, so all's well that ends well I guess.

TaylorAlexander 2y ago
Agreed. I think the short term profit motive is really blinding people to the risks. Like we could have continued this in a research setting for longer to really get a sense for how to deal with these things, but folks wanted to productize because of profit potential backed by incredible hype.