You can think of theorem provers as really crazy type checkers. It's not just a handful of tests that have to run, but more like a program that has to compile.
Yes exactly. There is this thing called the “Curry-Howard Isomorphism” which (as I understand it) says that propositions in formal logic are isomorphic to types. So the “calculus of constructions” is a typed lambda calculus based on this that makes it possible for you to state some proposition as a type and if you can instantiate that type then what you have done is isomorphic to proving the proposition. Most proof assistants (and certainly Lean) are based on this.
So although lean4 is a programming language that people can use to write “normal” programs, when you use it as a proof assistant this is what you are doing - stating propositions and then using a combination of a (very extensive) library of previous results, your own ingenuity using the builtins of the language and (in my experience anyway) a bunch of brute force to instantiate the type thus proving the proposition.
Technically, it isn't an isomorphism (the word is abused very often), and there is no fixed, general syntactic correspondence. However, in the case of Lean, we can establish a correspondence between its dependent type system and intuitionistic higher-order predicate logic.
You too can start your own space exploration company if you are lucky enough to somehow obtain capital and your luck compounds sufficiently from that capital.
That luck is out of grasp of the vast majority of the population. It isn't a matter of effort, but mostly luck. Playing the lottery is not a realistic strategy to put forth from a monopoly competition perspective. 90% of startups fail, for example. They worked hard, but were unlucky.
We should do both and it makes sense that different orgs have different focuses. It makes no sense to berate one set of orgs for not working on the exact type of thing that you want. PauseAI and ControlAI have each received less than $1 million in funding. They are both very small organizations as far as these types of advocacy non-profits go.
If it makes sense to handle all of these issues, then couldn't these organizations just acknowledge all of these issues? If reducing harm is the goal, I don't see a reason to totally segregate different issues, especially not by drawing a dividing line between the ones OpenAI already acknowledges and the ones it doesn't. I've never seen any self-described "AI safety" organizations that tackles any of the present-day issues AI companies cause.
If you've never seen it then you haven't been paying attention. For example Anthropic (the biggest AI org which is "safety" aligned) released a big report last year on metal well being [1]. Also here is their page on societal impacts [2]. Here is PauseAI's list of risks [3], it has deepfakes as its second issue!
The problem is not that no one is trying to solve the issues that you mentioned, but that it is really hard to solve them. You will probably have to bring large class action law suits, which is expensive and risky (if it fails it will be harder to sue again). Anthropic can make their own models safe, and PauseAI can organize some protests, but neither can easily stop grok from producing endless CSAM.
PauseAI's official proposal recommends[0]: "Only allow deployment of models after no dangerous capabilities are present." Their list of dangerous capabilities[1] does not include deepfakes, but it does include several unrealized ones that fit the description of this post here, including "a recursive loop of self-improvement, spinning rapidly out of control... called an intelligence explosion".
I appreciate you pointing out the Risks page though, as it does disprove my hyperbole about ignoring present-day harms completely, although I was disheartened that the page just appears to list things that they believe actions "could be mitigated by a Pause" (emphasis mine).
The amount of cattle required to maintain pasture is way fewer than we have right now. From a CO2 perspective factory farmed cattle tends to look a little better than "free-range" mostly due to reduced land use changes (but it is obviously worse from a cruelty perspective). Finally, we can still have farm animals without eating them!!
reply