casebash

They are still dedicated towards trying to advance humanity and I think that's more important than non-profit status. (For the record, I'm uncertain whether they are net positive or not given that I am skeptical of their approach to safety).

True, but the next logical step is to put these models in a standard reinforcement learning environment. See: https://sites.research.google/palm-saycan

(I mean like with a proper world model and not just RLHF which they are already doing).

Given that you've already updated your views recently, perhaps you'd be open to reconsidering whether AI could form an existential risk?

One of the best talks (but more technical talks) I've seen on this topic is Evan Hubringer's "How likely is deceptive alignment".

https://www.alignmentforum.org/posts/A9NxPTwbw6r6Awuwt/how-l...

It may also be worthwhile checking out the Rob Miles video on the Orthogonality Thesis - https://www.youtube.com/watch?v=hEUO6pjwFOo

Thanks, I really appreciate it!

Hmm... as much as I like John Wentworth, some of his posts there are kind of rants (I expect he'd admit that), so maybe not the best first introduction?

Is there any chance you could add some links to AI Safety Resources such as Joe Carlsmith's Power Seeking AI (https://forum.effectivealtruism.org/posts/ChuABPEXmRumcJY57/...) or Rob Miles Introduction to AI Safety (https://www.youtube.com/watch?v=pYXy-A4siMw)?

Perhaps you could link to a resource on the AI box experiment as well?

I know you vaguely gesture at it, but would likely be better to explicitly link people to resources if they haven't heard of it themselves.

I guess the universe is allowed to be crazy.

The universe is allowed to decide the Donald Trump will be president in 2016 despite all the reasons to think it was crazy that he would become president. The universe is allowed to decide that Volodymyr Zelenskyy will be elected president of Ukraine on the basis of having starring as the president in a comedy.

And I guess the universe is allowed to decide that a piece of fanfic that is desperately in need of an editor will be successful at recruiting talent for the rationalist or AI Safety communities.

I expect you are probably skeptical of AI Safety, but then your criticism would be a criticism of the final objective, not the method (distribution of HPMoR) used to achieve the objective.

On one hand the author seems to make fun of the notion that language models could be dangerous.

But then a few sentences down the author complains:

"The company “in charge” of protecting us from harmful AIs decided to let people use a system capable of engaging in disinformation and dangerous biases so they could pay for their costly maintenance."

So the author doesn't even present a coherent position.

What changes?

Even if there are things that quantitative measurement can't capture, perhaps we could at least use it to rule out donating to some charities?

Your assumption seems to be that either it is a pure emotional value judgement (like donating to a dog shelter because you felt like it) or a perfectly objective judgement (ie. meta-reviews of randomised controlled trials). Perhaps that's a false dichotomy and there's a difference between supporting a cause based that is now supported by a significant body of philosophical and technical literature and just donating to whatever you feel like? There's nothing wrong with the later, just that you are kidding yourself if you think it is likely to be effective.

China still scores near the top of the international maths olympiads. These competitions are designed to force you to require creativity and not just memorise formulas So even if the system doesn't produce high average creativity, it seems to produce high creativity at the top level.

Have you considered adding an advanced run option that doesn't obliterate the code in the REPL?

In particular, it would be great update a function or class select it and press a button to run the updated version in the REPL so that you wouldn't have to copy and paste it there manually.

Should we be eating iodised salt as well? Or is it only important for children/pregnant women?

I wouldn't say that innovation in Desktop is dead, but most of it seems to be driven by features or design patterns copied from mobile or tablet. Take for examples Windows 8 and Windows 10, Windows 8 was all about moving to an OS that could run on a whole host of devices, while Windows 10 was all about fixing up all the errors made in this transition.

"probability a non-DA will assault someone in a given encounter" - So this assumes that if someone has enough encounters that they will eventually have near 100% chance of committing assault.

"probability a non-DA will assault someone in a given encounter" - so according to this model, if someone has enough sexual encounters there is a near 100% chance that they will assault someone. That's kind of dodgy.

Is there any reason to complete these puzzles if your resume/contacts are strong enough to get you an interview anyway? Apart from fun, that is?

I hate Twitter as well. It is occupying a place that could very easily be occupied by much better products. It's primary purpose is to hold back innovation.

I'm sure some people prefer Twitter as it is and it is valid for them to do so. However, the existence of Twitter is still massively harmful in terms of opportunity cost.

Facebook's strategy to create separate apps is a good one, but Rooms seem rather niche. Facebook groups are better for many purposes since they have access to notifications and the newsfeed. Reddit probably works better for large groups and allows anonymity too (though not on a per Reddit basis). So Rooms work for groups where anonymity is valuable (ie. sharing secrets, controversial opinions, ect) and where the groups aren't too large. Here, the discovery mechanism being overly complex actually does provide an advantage. It makes it more likely to be "underground" and hence cool.

I think that it is strange that she is saying that there is too much focus on understanding at school. From what I could see, most people learned maths by just memorising the process. Except for a few who invested an exceptional amount of effort, they scored poorly. In contrast, everyone who I knew who focused on gaining a deep understanding was able to pick up maths quickly.

There is one area, though, where there is probably too much focus on understanding at the cost of memorisation and that is in the Olympiads. I remember that many of us held the rather uncharitable attitude that the people who tried to succeed in maths by memorising it were 'stupid'. But clearly, understanding combined with targeted memorisation of the key building blocks will lead to the most success.