4 ms·

Link to the article: https://www.qualcomm.com/news/releases/2023/07/qualcomm-works-with-meta-to-enable-on-device-ai-applications-usi https://www.qualcomm.com/ne

by pera 3y ago

Link to the article: https://www.qualcomm.com/news/releases/2023/07/qualcomm-works-with-meta-to-enable-on-device-ai-applications-usi https://www.qualcomm.com/news/releases/2023/07/qualcomm-work...

qwertox 3y ago
> "We applaud Meta’s approach to open and responsible AI and are committed to driving innovation and reducing barriers-to-entry for developers of any size by bringing generative AI on-device,” Can someone explain to me why Meta's approach is responsible? I mean, I applaud Meta for "open sourcing" the models, but don't they contain potentially harmful data which can be accessed without some kind of filter? Let's say retrieve instructions on how to efficiently overthrow a government?
- circuit10 3y ago
  There are risks to LLMs but I don’t think that’s a major one, it’s probably easier to find that through Google
- sangnoir 3y ago
  > Let's say retrieve instructions on how to efficiently overthrow a government? Your license to use Llama can be revoked if Meta investigates and deems your action to be against the code of conduct[1]. I suppose if you continue to use the model without a license, you may be open to subsequent legal action in uncharted case-law territory. 1. https://github.com/facebookresearch/llama/blob/main/CODE_OF_CONDUCT.md https://github.com/facebookresearch/llama/blob/main/CODE_OF_...
  - gmt2027 3y ago
    Are you saying that the threat of civil legal action is an effective deterrent against anonymous actors that conspire to overthrow governments?
    - sangnoir 3y ago
      I never made a claim on efficacy.
- stavros 3y ago
  Would the major obstacle in overthrowing a government be that you just don't know how? Even for things like making a bomb, you can probably find the instructions easily.
- throwuwu 3y ago
  > instructions on how to efficiently overthrow a government lol the examples people give for AI safety are always so ridiculous. Here I am thinking I’m bad at estimating when I misjudge a week’s worth of work for a month and this guy thinks he can overthrow a government by following a few thousand word plan written by an LLM.

by pera 3y ago

Link to the article: https://www.qualcomm.com/news/releases/2023/07/qualcomm-works-with-meta-to-enable-on-device-ai-applications-usi https://www.qualcomm.com/news/releases/2023/07/qualcomm-work...

qwertox 3y ago

> "We applaud Meta’s approach to open and responsible AI and are committed to driving innovation and reducing barriers-to-entry for developers of any size by bringing generative AI on-device,” Can someone explain to me why Meta's approach is responsible? I mean, I applaud Meta for "open sourcing" the models, but don't they contain potentially harmful data which can be accessed without some kind of filter? Let's say retrieve instructions on how to efficiently overthrow a government?

circuit10 3y ago
There are risks to LLMs but I don’t think that’s a major one, it’s probably easier to find that through Google
sangnoir 3y ago
> Let's say retrieve instructions on how to efficiently overthrow a government? Your license to use Llama can be revoked if Meta investigates and deems your action to be against the code of conduct[1]. I suppose if you continue to use the model without a license, you may be open to subsequent legal action in uncharted case-law territory. 1. https://github.com/facebookresearch/llama/blob/main/CODE_OF_CONDUCT.md https://github.com/facebookresearch/llama/blob/main/CODE_OF_...
- gmt2027 3y ago
  Are you saying that the threat of civil legal action is an effective deterrent against anonymous actors that conspire to overthrow governments?
  - sangnoir 3y ago
    I never made a claim on efficacy.
stavros 3y ago
Would the major obstacle in overthrowing a government be that you just don't know how? Even for things like making a bomb, you can probably find the instructions easily.
throwuwu 3y ago
> instructions on how to efficiently overthrow a government lol the examples people give for AI safety are always so ridiculous. Here I am thinking I’m bad at estimating when I misjudge a week’s worth of work for a month and this guy thinks he can overthrow a government by following a few thousand word plan written by an LLM.