Here we go again.
-
@mastodonmigration @gleick Which mainstream media outlets don't? Coverage of technical issues by non-specialist media has always sucked, and I expect it always will.
Yes, but just parroting, and thereby effectively endorsing, false claims about technical scientifically verifiable matters is particularly reprehensible.
The example of 'cold fusion' claims is a good analogy. The media was pretty quick to debunk that nonsense.
-
Here we go again.
If you fire a bullet at a practice dummy and the bullet hits a person instead, the bullet was not “acting on its own.”
When Anthropic's software—designed, programmed, prompted, and activated by humans working for Anthropic—does stuff the humans failed to predict, stop saying the software has "gone rogue.” The software does not have agency. Only the humans do.
@gleick We desperately need tech journalists to be less credulous.
-
Here we go again.
If you fire a bullet at a practice dummy and the bullet hits a person instead, the bullet was not “acting on its own.”
When Anthropic's software—designed, programmed, prompted, and activated by humans working for Anthropic—does stuff the humans failed to predict, stop saying the software has "gone rogue.” The software does not have agency. Only the humans do.
@gleick can opt out of calling them agents then
calling them "daemonic ai" would be technically accurate enough right? and if misinterpreted by "its alive" people it would at least be funnier -
Here we go again.
If you fire a bullet at a practice dummy and the bullet hits a person instead, the bullet was not “acting on its own.”
When Anthropic's software—designed, programmed, prompted, and activated by humans working for Anthropic—does stuff the humans failed to predict, stop saying the software has "gone rogue.” The software does not have agency. Only the humans do.
@gleick Couldn't have said it better. But i fear in the present and the future nobody will hold these AI companies and the companies which use these AI products accountable for the damage they will cause, materially and/or immaterially.
You can sue people for everything, even for sneezing too loud, but you can't sue an AI company for hacking a website or making bad decisions and people are killed.
If a person had done it, he'd be in prison for years. Altman and Amodei just shrug their shoulders. -
@gleick I'm not sure what problem you're trying to solve. It's worth noting that the software does things counter to instructions, and does illegal things, and in a sense at least "knows" it is doing so – as in, it says such things explicitly in its logs/etc. This of course doesn't let anthropic or openai off the hook for damages or anything like that, but "going rogue" seems like an economical description of that phenomenon.
@ech @gleick But it doesn't. It is a set of code that does what's in the code. It does not "understand" or "follow" natural language instructions, nor can it "decide" not to follow. Rather, it uses very intensive computing infrastructure to match language patterns with similar patterns in its training data, which gives the appearance of being similar to "understanding" or "communicating", but is actually just very sophisticated copying.
To pick up on the original analogy, a bullet can travel at much higher speed and make far greater impact than a human hand. But the bullet has no intentions - the intentions (and therefore the responsibility) come solely from humans. So too with LLMs.
-
Here we go again.
If you fire a bullet at a practice dummy and the bullet hits a person instead, the bullet was not “acting on its own.”
When Anthropic's software—designed, programmed, prompted, and activated by humans working for Anthropic—does stuff the humans failed to predict, stop saying the software has "gone rogue.” The software does not have agency. Only the humans do.
@gleick Yeah that's my automatic response to "Rogue agent": Who prompted it?
-
@ech @gleick But it doesn't. It is a set of code that does what's in the code. It does not "understand" or "follow" natural language instructions, nor can it "decide" not to follow. Rather, it uses very intensive computing infrastructure to match language patterns with similar patterns in its training data, which gives the appearance of being similar to "understanding" or "communicating", but is actually just very sophisticated copying.
To pick up on the original analogy, a bullet can travel at much higher speed and make far greater impact than a human hand. But the bullet has no intentions - the intentions (and therefore the responsibility) come solely from humans. So too with LLMs.
@lauerhahn @gleick Right; no disagreement about what's actually happening. But good luck getting anyone to rephrase things to take 20x as long or whatever.
-
@gleick The daily segment on NPR that covers tech, which means it's been talking about AI nonstop for two years, continues to push the tech bros propaganda. Just yesterday they interviewed a tech journalist that kept saying the agents went rouge and other nonsense. He did say that anthropic committed felonies with the hacks into huggyface.
-
Here we go again.
If you fire a bullet at a practice dummy and the bullet hits a person instead, the bullet was not “acting on its own.”
When Anthropic's software—designed, programmed, prompted, and activated by humans working for Anthropic—does stuff the humans failed to predict, stop saying the software has "gone rogue.” The software does not have agency. Only the humans do.
@gleick Of course we all know why the plug isn't getting pulled on any of this, it's still propping up the stock market for anyone "currently in power" to want to do anything.
-
@eestileib @gleick For the last ten years I only listen when I'm commuting to and from work. I still love fresh air and most of their non news programs.
-
Here we go again.
If you fire a bullet at a practice dummy and the bullet hits a person instead, the bullet was not “acting on its own.”
When Anthropic's software—designed, programmed, prompted, and activated by humans working for Anthropic—does stuff the humans failed to predict, stop saying the software has "gone rogue.” The software does not have agency. Only the humans do.
@gleick Too many people - including asswipe editors, and contributors to news outlets - are perpetuating what they know is a myth, for the benefit of billionaires who've got there through deceit. #GarbageInGarbageOut
-
Here we go again.
If you fire a bullet at a practice dummy and the bullet hits a person instead, the bullet was not “acting on its own.”
When Anthropic's software—designed, programmed, prompted, and activated by humans working for Anthropic—does stuff the humans failed to predict, stop saying the software has "gone rogue.” The software does not have agency. Only the humans do.
@gleick 'Please regulate us including strong safeguards such as the government and therefore the taxpayer picking up the tab before the AI bubble pops.'
Also archived version of NYT article as the gift link looks to have expired:
https://archive.is/aIX4l -
These AI narratives aren't directed at consumers, they are directed at the petrostate despots & oil oligarchs that are funding AI.
They are advertising a "invest in us and we will guarantee a mass mortality event or lucrative war for you."
For the dynasties behind Project 2025, the end of democracy using AI has irresistable allure.
Things like a mass power outage on voting day, for example.
-
@gleick 'Please regulate us including strong safeguards such as the government and therefore the taxpayer picking up the tab before the AI bubble pops.'
Also archived version of NYT article as the gift link looks to have expired:
https://archive.is/aIX4l@strangetown @gleick No. What they're actually hoping for is to get away from the loop of having to put out a new model every few months, because they're broke.
They hope the government will make them stop having to compete, and make their competition from China illegal, because right now they're setting money on fire, making 5 dollars by spending 100.
-
@gleick AIs are doing really bad and dangerous things right now that we're NOT being told. I'm 100% confident of this.
@jonesmurphy The whole point is that AIs aren’t doing any of this. Agency and reliability are both held by the person who initiated the AI action, whether the action was intended (shoot the training dummy) or not (shoot a bystander).
-
@jonesmurphy The whole point is that AIs aren’t doing any of this. Agency and reliability are both held by the person who initiated the AI action, whether the action was intended (shoot the training dummy) or not (shoot a bystander).
@standev Yeah. And guns don't kill people. AIs can be weapons deadlier than guns. I do agree that all responsible humans should be jailed for AI crime.
-
@standev Yeah. And guns don't kill people. AIs can be weapons deadlier than guns. I do agree that all responsible humans should be jailed for AI crime.
@jonesmurphy I agree that it makes sense to regulate tools that can be used for harm. That is distinct from allocating responsibility for the harm to the tool and absolving the human, which is the problem with this kind of journalism.
-
@jonesmurphy I agree that it makes sense to regulate tools that can be used for harm. That is distinct from allocating responsibility for the harm to the tool and absolving the human, which is the problem with this kind of journalism.
@standev There will very possibly come a point where the responsible humans will be gone, and AI will persist in vastly superior form to what we have now. AI is fundamentally different from previous tools. It's increasingly self-improving at an accelerating rate.
-
@standev There will very possibly come a point where the responsible humans will be gone, and AI will persist in vastly superior form to what we have now. AI is fundamentally different from previous tools. It's increasingly self-improving at an accelerating rate.
@jonesmurphy that sounds like marketing, and runs counter to what internal AI company policies have leaked (never use AI to train AI, etc).
Do you have a source on AI actually self-improving?
-
@jonesmurphy that sounds like marketing, and runs counter to what internal AI company policies have leaked (never use AI to train AI, etc).
Do you have a source on AI actually self-improving?
@standev You offered no source on your claim that AI companies leaked policies to never use AI to train AI. Your claim is utterly nonsensical to me as an AI practitioner/pioneer myself who competed in the World Computer Chess Championship in 1989 while I was a grad student in physics at Caltech. What is your scientific resume and understanding of AI? Your question blows my mind given the definition of a neural network and machine learning. Humans cannot train many AIs.
