321–330 of 409 posts
Earlier quoted context omitted.
That is precious time that can be used to work on alignment.
But alignment is always going to rely on cooperation of users though? What benefit does the delay offer other than the direct one of a delay?
Earlier quoted context omitted.
Terrorists already have all the information they need to build some heinous shit with ~no external guidance aside from what's already on the internet.
Engineered viruses could cause far more deaths than conventional weapons. Even more than nuclear weapons, and they are easier to manufacture.
Because Meta is releasing their models to the public, I consider them the most ethical company doing AI at scale. Keeping AI models closed under the guise of “ethics”, is I think the most unethical stance as it makes people more dependent on the arbitrary decisions, goals, and priorities of big companies, instead being allowed to define “alignment” for themselves.
Earlier quoted context omitted.
Let's be honest here, a lot of HN users have a conflict of interest on this topic. AI entrepreneurs trying to get rich from LLMs benefit from LLMs being open source. But the downside risk from e.g. bioweapons is spread across all of society. It's the same sort of asymmetrical cost/benefit that tobacco companies and greenhouse gas emitters face. Of course if you went to an online forum for oil companies, they'd be hop…
> But the downside risk from e.g. bioweapons is spread across all of society. It gets old hearing about these "risks" in the context of AI. It's just an excuse used by companies to keep as much power as possible to themselves. The real risk is AI being applied in decision making where it affects humans.
https://www.stat.berkeley.edu/~aldous/157/Papers/yudkowsky.p...
https://intelligence.org/files/AIPosNegFactor.pdf
I am concerned with AI companies keeping all the power to themselves. The recent announcement from the OpenAI board was encouraging on that front, because it makes me believe that maybe they aren't focused on pursuing profit at all costs.
Even so, in some cases we want power to be restricted. For example, I'm not keen on democratizing access to nuclear weapons.
>The real risk is AI being applied in decision making where it affects humans.
I agree. But almost any decision it makes can affect humans directly or indirectly.
In any case, the more widespread the access to these models, the greater the probability of a bad actor abusing the model. Perhaps the current generation of models won't allow them to do much damage, but the next generation might, or the one after that. It seems like on our current path, the only way for us to learn about LLM dangers is the hard way.
AI safety is just a rent seeking cult/circus, leeching on the work done by others. Good on Meta for cleaning shop.
To me, the greatest apocalypse scenario is not some AGI global extinction event but a corporation with an extensive data hoard replete with ample ML and GPU power being able to monopolize a useful service that cannot be matched by the public... that is the true, (and likely imo) AI nightmare we're heading towards.
Earlier quoted context omitted.
Let's be honest here, a lot of HN users have a conflict of interest on this topic. AI entrepreneurs trying to get rich from LLMs benefit from LLMs being open source. But the downside risk from e.g. bioweapons is spread across all of society. It's the same sort of asymmetrical cost/benefit that tobacco companies and greenhouse gas emitters face. Of course if you went to an online forum for oil companies, they'd be hop…
It is worth calling out the motivations of most entrepreneurs here. But I think that analogy you used is very uncharitable - drilling and burning fossil fuels necessarily harms the environment, but the track record on big companies handling alignment/safety in house, rather than open source with the whole research community working on it is still very much up in the air. Sidney (bing assistant) was easy to prompt inj…
Yep, Microsoft did a terrible job, and they should've been punished.
I'm not claiming that Big AI rocks at safety. I'm claiming that Big AI is also a big target for regulators and public ire. There's at least a chance they will get their act together in response to external pressure. But if cutting-edge models are open sourced indefinitely, they'll effectively be impossible to control.
>research that people have been able to do on forcing the output of llama to confirm to certain rules will likely prove invaluable in the future.
You may be correct that releasing llama was beneficial from the point of view of safety. But the "conform to certain rules" strategy can basically only work if (a) there's a way to enforce rules that can't be fine-tuned away, or (b) we stop releasing models at some point.
Earlier quoted context omitted.
expert?! can’t do math why should suggest weapon instructions better? at the first hallucination you explode on the spot
That's why we're not already dead. If anyone releases all the weights of a model that does everything perfectly (or at least can use the right tools which I suspect is much easier), that model is far too valuable to make it disappear, and dangerous enough to do all the things people get worried about. The only way to prevent that is to have a culture of "don't release unless we're sure it's safe" well before you reac…
You mean a complete hypothetical outside of scifi? Lets start worrying about alien invasions too?
Our planet is actually, not hypothetically, becoming uninhabitable due to pollutiom. I am so tired of ML people thinking they are god and have created something of infinite power. The hubris.
The birds eye view is that we need tons of major breakthroughs to allow us to overcome this climate disaster while also figuring out how to make 8 Billion+ comfortable and happy without turning the earth into a toxic waste dump, and we need this ASAP. This nonsense about AI safety is going to have a negative net affect on the lives of Billions of people, by slowing down the progress thay could be made.
AI X-risk is a complete sham being used to try and control a new, powerful tool. Science requires the scientific method, which requires physical embodiment, trial and error, and disciplined observation and measurment. AI has 0 ability to do any of that, we don't even have online learning(I think that's the term, where the model learns from its usage) in any of these large models.
Earlier quoted context omitted.
> But the downside risk from e.g. bioweapons is spread across all of society. It gets old hearing about these "risks" in the context of AI. It's just an excuse used by companies to keep as much power as possible to themselves. The real risk is AI being applied in decision making where it affects humans.
Concern with AI risk actually originated decades ago, way before current AI hype. Here are some of Eliezer Yudkowsky's papers from 2006/2008: https://www.stat.berkeley.edu/~aldous/157/Papers/yudkowsky.p... https://intelligence.org/files/AIPosNegFactor.pdf I am concerned with AI companies keeping all the power to themselves. The recent announcement from the OpenAI board was encouraging on that front, because it makes…
Earlier quoted context omitted.
All of the information AI regurgitates is either already available online as part of its corpus (and therefore the AI plays no particular role in access to that information), or completely made up (which is likely to kill more terrorists than anyone else!) Reiterating other comments, terrorists can't make bioweapons because they lack the facilities and prerequisites, not because they're incompetent.
The "all the info is already online" argument is also an argument against LLMs in general. If you really believe that argument, you shouldn't care one way or another about LLM release. After all, the LLM doesn't tell you anything that's not on Google. Either the LLM is useful, in which case it could be useful to a terrorist, or it's useless, in which case you won't mind if access is restricted. Note: I'm not saying i…
No it isnt. That's like saying, "You can walk" is an argument against cars.