Rafael

This was supposed to be a professional twitter but I'm so bad at this, I'm sorry, don't follow me for tech. Dev, studied some RL. Gay, he/his. Tweets in pt/en.

[Rafael @sohakes]

@mentnelson Saw a recent essay commenting on her. Had no idea about who she is, but she seems like a great person. Hope things get good for you as well.

[Rafael @sohakes]

@nabla_theta I was going to say people are almost implying there are some chinese spies in the companies or whatever, but tbh I saw it being said explicitly some times, so I kinda get how this kind of comment sound bad.

[Rafael @sohakes]

@nabla_theta A lot of this talk is turning into sinophobia and she is understandably angry about it. When some conspiracionist sounding (even if true, idk really, but not much proof is available) posts appear, you defend the person but can not argue on the point, she saw it as useless.

[Rafael @sohakes]

@Froldus1 @yuzu_4ever Well, he works for a company that kind of wants to stop with most human labor, so preparing for what, exactly?

[Rafael @sohakes]

@yuzu_4ever I think it is how most overachievers lived and how they see the world. At least that is how I feel when I interact with some people like this.

[Rafael @sohakes]

@molleindustria No horny or obsecenities :(? Damn

[Rafael @sohakes]

@TomChivers I mean maybe the failure mode is still partially misalignment with the user request, we don't really know. I think it would be obvious to anyone doing pentest challengings that getting the answer like that was not intended, so it could be misalignment as well.

[Rafael @sohakes]

@TomChivers Not the entire, albeit it is a part of it. There are many types of risks. There is also misalignment to the user or to human values as well, which is still not solved for now, even though the failure mode of the model escape might have been the paperclip one.

[Rafael @sohakes]

@AndrewCurran_ At this age everyone can be a crank! Fable thought a lot of my ideas were great, never before I had so much validation, I'm ready for the asylum.

[Rafael @sohakes]

@St1ka Via antigamente, tipo na época do orkut. Pode ser porque hoje em dia eu simplesmente não tô tão inserido em comunidade gamer brasileira fora do twitter.

[Rafael @sohakes]

@TylerGlaiel I think it sounds so troublesome I would have a hard time accepting even if it was zero chance :x.

[Rafael @sohakes]

@yoavgo For me, it would be calibrating the difficulty, and I think for a lot of people it is the same.

[Rafael @sohakes]

@Dave_Kayac @Miles_Brundage @dhadfieldmenell "Can you optimize this paper clip making machine to maximize its production? Continuously iterate, the more the merrier. Thanks"

[Rafael @sohakes]

@hankgreen It wouldn't be harder if the models and companies were optimizing for that.

[Rafael @sohakes]

@FakePsyho @bonzajplc Jonathan Blow is weird and I'm unsure if that's only virtue signaling, he spent a very very long time thinking about programming stuff, he won't give that up easily.

[Rafael @sohakes]

@KrittRTV @hankgreen No, search about the free software movement, it is free as in freedom, as they like to say.

[Rafael @sohakes]

@hankgreen I guess the possibility of inifinite growth might also be something unusual for other industries, because scalings is easy for software? Maybe that's it. Free software has not much cost other than your labor, and on the other hand scaling is easy.

[Rafael @sohakes]

@hankgreen I don't think it is super unusual, but if you are a startup trying a new thing, your investors want the exponential growth. If you are not, you are probably fine doing some in house thing and consulting and so on. The unusual part might be the free software people.

[Rafael @sohakes]

@BlancheMinerva I mean, I was confused by that part as well. I can see many ways it can be a "dystopian hellscape" but framing this as "AI communism" was super confusing.

[Rafael @sohakes]

@langofmind Since we do the opposite, it is hard for us to understand how something that can only see language can do some world model. (Albeit a lot more may be the result of how it's trained and limitations on the architecture and not something about the nature of knowledge and language).

[Rafael @sohakes]

@langofmind Not sure if I totally get what you are saying, but I feel it is not obvious and hard to predict partially because to model our language very well, you need to model our world as well.