4 ms·

> Now suppose that it's open source and runs on your own device. I agree that AI could be a massive asset to people and help improve our lives if it worked for

> Now suppose that it's open source and runs on your own device. I agree that AI could be a massive asset to people and help improve our lives if it worked for us and our interests. Sadly that's not how things seem to work out. Every piece of technology on offer today seems to work for someone else. Our phones, our cars, our TVs, our computers, our appliances, even our medical devices. All of our tech works for other someone other than the person who paid for it and mostly so that companies can exploit us while delivering only a fraction of what those devices are capable of. Features are often locked or paywalled off to extract more money or encourage additional purchases or to prevent you from blocking ads or taking back control of your hardware and the software it runs. Our usage is monitored and data collected to be used against us later. Even now AI is largely controlled by private companies. Even as the hardware requirements to run AI locally go down, we'll never compete with companies like facebook, google, and microsoft in terms of the size of the datasets they have to train with. Worse I think we're starting to see efforts by corporations to push for laws that limit the ability for people to use AI. They'll say it's too dangerous to let regular people use AI, and that only they can be trusted with such power.

AnthonyMouse 3y ago
> Every piece of technology on offer today seems to work for someone else. You can buy a commodity PC and install Debian on it. Some people do. > Even as the hardware requirements to run AI locally go down, we'll never compete with companies like facebook, google, and microsoft in terms of the size of the datasets they have to train with. Most of these datasets are public. > Worse I think we're starting to see efforts by corporations to push for laws that limit the ability for people to use AI. They'll say it's too dangerous to let regular people use AI, and that only they can be trusted with such power. Of course they will. But the "AI is bad" rhetoric only feeds into that. What do you expect an "AI is bad" bill to look like? It will be the one that restricts access to only large institutions.
- autoexec 3y ago
  > You can buy a commodity PC and install Debian on it. But your motherboard will still be full of chips that are closed and proprietary and some of them will have their own OS with network/wireless access that you'll have no control over. You'll just have to trust that the Intel Management Engine backdoor and radio chipsets you have no access to aren't taking advantage of you. > Most of these datasets are public. Most of everything Google has indexed and cached is "public". Libraries are filled with the same books Google has already scanned. Good luck using the entire internet and your local library to train your own personal AI though. Microsoft has access to an incredible amount of non-public data including the emails of corporations using outlook and the files on people's hard drive. If they were to fully leverage their access to train AI you'd have a hard time keeping up. Facebook has every post and "private" message ever sent on the platform including ones that were deleted years ago. We can't expect a few publicly released datasets to compare to what they're capable of getting. Even if we had that kind of access to raw data, could we realistically process it? https://fortune.com/2024/04/04/ai-training-costs-how-much-is-too-much-openai-gpt-anthropic-microsoft/?taid=660ed6f0ab6464000175d1ad https://fortune.com/2024/04/04/ai-training-costs-how-much-is... > But the "AI is bad" rhetoric only feeds into that. What do you expect an "AI is bad" bill to look like? It will be the one that restricts access to only large institutions. There are a lot of different versions of "AI is bad". One is "AI is bad when it's only controlled by corporations", but that's not what the laws will be about. They're more likely to say that "AI is bad" because of terrorists and spammers and pedos or whatever other boogeymen they can summon up to scare the ignorant and provide some justification for keeping that power out of the public's hands.
  - AnthonyMouse 3y ago
    > You'll just have to trust that the Intel Management Engine backdoor and radio chipsets you have no access to aren't taking advantage of you. And that's dumb and we should fix it, but there is no evidence that they are actually doing this. Conversely, there is plenty of evidence that e.g. John Deere or Apple is doing this. > Good luck using the entire internet and your local library to train your own personal AI though. Presumably you would use Common Crawl and Project Gutenberg etc. > Microsoft has access to an incredible amount of non-public data including the emails of corporations using outlook and the files on people's hard drive. If they were to fully leverage their access to train AI you'd have a hard time keeping up. Then they'd set their credibility on fire, lose their customers and get sued because their model would be capable of emitting all of their customers' confidential private information. > Even if we had that kind of access to raw data, could we realistically process it? Yes. Training is a fixed cost that can be amortized over everyone using the model. There are various methods of spreading the cost or funding the model: https://news.ycombinator.com/item?id=39964361 https://news.ycombinator.com/item?id=39964361 > There are a lot of different versions of "AI is bad". One is "AI is bad when it's only controlled by corporations" That one isn't really "AI is bad" at all, it's really "market concentration is bad" and has nothing to do with AI except insofar as AI is another thing that could be monopolized by robber barons. > They're more likely to say that "AI is bad" because of terrorists and spammers and pedos or whatever other boogeymen they can summon up to scare the ignorant and provide some justification for keeping that power out of the public's hands. These justifications are all horse manure and should be repeated only sarcastically in the tone of Helen Lovejoy. AI is bad in the same way that electricity is bad. "Electricity could be used by pedos" is an argument so weak that the only thing it proves is that the person offering it is arguing in bad faith.

AnthonyMouse 3y ago

> Every piece of technology on offer today seems to work for someone else. You can buy a commodity PC and install Debian on it. Some people do. > Even as the hardware requirements to run AI locally go down, we'll never compete with companies like facebook, google, and microsoft in terms of the size of the datasets they have to train with. Most of these datasets are public. > Worse I think we're starting to see efforts by corporations to push for laws that limit the ability for people to use AI. They'll say it's too dangerous to let regular people use AI, and that only they can be trusted with such power. Of course they will. But the "AI is bad" rhetoric only feeds into that. What do you expect an "AI is bad" bill to look like? It will be the one that restricts access to only large institutions.

autoexec 3y ago
> You can buy a commodity PC and install Debian on it. But your motherboard will still be full of chips that are closed and proprietary and some of them will have their own OS with network/wireless access that you'll have no control over. You'll just have to trust that the Intel Management Engine backdoor and radio chipsets you have no access to aren't taking advantage of you. > Most of these datasets are public. Most of everything Google has indexed and cached is "public". Libraries are filled with the same books Google has already scanned. Good luck using the entire internet and your local library to train your own personal AI though. Microsoft has access to an incredible amount of non-public data including the emails of corporations using outlook and the files on people's hard drive. If they were to fully leverage their access to train AI you'd have a hard time keeping up. Facebook has every post and "private" message ever sent on the platform including ones that were deleted years ago. We can't expect a few publicly released datasets to compare to what they're capable of getting. Even if we had that kind of access to raw data, could we realistically process it? https://fortune.com/2024/04/04/ai-training-costs-how-much-is-too-much-openai-gpt-anthropic-microsoft/?taid=660ed6f0ab6464000175d1ad https://fortune.com/2024/04/04/ai-training-costs-how-much-is... > But the "AI is bad" rhetoric only feeds into that. What do you expect an "AI is bad" bill to look like? It will be the one that restricts access to only large institutions. There are a lot of different versions of "AI is bad". One is "AI is bad when it's only controlled by corporations", but that's not what the laws will be about. They're more likely to say that "AI is bad" because of terrorists and spammers and pedos or whatever other boogeymen they can summon up to scare the ignorant and provide some justification for keeping that power out of the public's hands.
- AnthonyMouse 3y ago
  > You'll just have to trust that the Intel Management Engine backdoor and radio chipsets you have no access to aren't taking advantage of you. And that's dumb and we should fix it, but there is no evidence that they are actually doing this. Conversely, there is plenty of evidence that e.g. John Deere or Apple is doing this. > Good luck using the entire internet and your local library to train your own personal AI though. Presumably you would use Common Crawl and Project Gutenberg etc. > Microsoft has access to an incredible amount of non-public data including the emails of corporations using outlook and the files on people's hard drive. If they were to fully leverage their access to train AI you'd have a hard time keeping up. Then they'd set their credibility on fire, lose their customers and get sued because their model would be capable of emitting all of their customers' confidential private information. > Even if we had that kind of access to raw data, could we realistically process it? Yes. Training is a fixed cost that can be amortized over everyone using the model. There are various methods of spreading the cost or funding the model: https://news.ycombinator.com/item?id=39964361 https://news.ycombinator.com/item?id=39964361 > There are a lot of different versions of "AI is bad". One is "AI is bad when it's only controlled by corporations" That one isn't really "AI is bad" at all, it's really "market concentration is bad" and has nothing to do with AI except insofar as AI is another thing that could be monopolized by robber barons. > They're more likely to say that "AI is bad" because of terrorists and spammers and pedos or whatever other boogeymen they can summon up to scare the ignorant and provide some justification for keeping that power out of the public's hands. These justifications are all horse manure and should be repeated only sarcastically in the tone of Helen Lovejoy. AI is bad in the same way that electricity is bad. "Electricity could be used by pedos" is an argument so weak that the only thing it proves is that the person offering it is arguing in bad faith.