Bachelor in general and applied physics. AI safety researcher wannabe. Interested in agent foundations.
Email: roman.malov27@gmail.com
GitHub: https://github.com/RomanMalov
TG channel (in Russian): https://t.me/healwithcomedy
Roman Malov
Karma: 91
I think are possible (and smth like this is already being used in complex functions visualizations). Not sure if you could display i.e. 5d hypercube this way (by the same reason there are no function which looks like a square)
and
Shouldn’t the second singularity be at the point ?
It would be interesting if any of them decided to (instrumentally) stream games (using a vtuber avatar for example) to earn money from donations. They need to figure out how to actually be a good streamer in order for this to work.
Why not just save them to an offline hard drive?
I am a bit confused. If the question is, ‘Will this alignment paradigm work with superintelligence?’ is the recommendation from the tweet to try it and see if it works?
I meant to imply that we do not have a robot capable of performing tasks of a similar level of difficulty to the ‘saving grandma’ task, with safety properties comparable to those that a human firefighter can provide when performing ‘saving grandma’ task.
Thanks for pointing that out, I will adjust the post.
I recently prepared an overview lecture about research directions in AI alignment for the Moscow AI Safety Hub. I had limited time, so I did the following: I reviewed all the sites on the AI safety map, examined the ‘research’ sections, and attempted to classify the problems they tackle and the research paths they pursue. I encountered difficulties in this process, partly because most sites lack a brief summary of their activities and objectives (Conjecture is one of the counterexamples). I believe that the field of AI safety would greatly benefit from improved communication, and providing a brief summary of a research direction seems like low-hanging fruit.
So, is a random variable in the sense that it is drawn from a distribution of functions, and the expected value of those functions at each point is equal to . Am I understanding you correctly?
I’ve read it as a part of Agents Foundation course, and I consider this post really effective and clarifying. It got me thinking, can this generalize to other failure modes? Like if programers notice that AI spend too much resources on self-preservation, and then train against such behavior, this failure mode would still arise because self-preservation is an instrumental goal and is a fact about the world and ways in which goal can be achieved in this world.
I’m not a native speaker, can someone please explain the meaning of “Hell is wasted on the evil” in simpler terms?
Thank you, that seems to be the clarification I needed. And reminded me of a good video, which also touches the subject.
Thank’s for your answer, I will read linked post.
I told in the text that I’m going to try to convey the “process” in the comments, and I’ll try to do it now.all sophisticated-enough minds
I think that the recursive buck is passed to the word “enough”. You need to have stratification of sophistication of minds, and have a cutoff for when they reach acceptable level off sophistication.
So in the universe with only bosons (so Pauli principle doesn’t apply), everything is the Same?
When I imagine a room full of photons, I see a lot of things that can be Different. For example, the coordinates of photons, wavelength, polarization, their number.
Or are you saying that Pauli principle is sufficient, but not necessary?
If you read further, you can see how this is also passing the recursive buck.
You: “There are no clear separation between objects, I only use this to increase my utility function”
Me: “How are you deciding on where to stop dividing reality?”
You: “Well, I calculate my marginal utility from creating an additional concept and then Compare it to zer… ah, yeah, there is the recursive buck. It even capitalized as I said it.”
So yeah, while this is a desirable point to stop, this method still relies on your ability to Differentiate between usefulness of two models, and as far as I can tell, in the end, we can only feel it.
Sebz n gval fcbg ba gur raq bs Uneel’f jnaq, n phovp zvyyvzrgre bs napube, fgergpurq bhg n guva yvar bs Genafsvtherq fcvqre-fvyx.
sebz gur puncgre 114
Or if I’d—if I’d only gone with—if, that night -
I’m guessing he is talking about the night he lost his potential phoenix.
I think that’s intended author’s choice. Like what Harry saw was too terrible to acknowledge. Or maybe it’s just to create more suspense.
Snape told him that he wanted to check if Harry resembled his father, and the test consisted of stopping bullies, so that might be the reason for Harry’s guess.
safety always remains ahead
When was it ever ahead? I mean, to be sure that safety is ahead, you need to first make advancement there compatible with capabilities. And to do that, you shouldn’t advance the capabilities.
Rule and Example
Rules can generate examples. For instance: DALLE-3 is a rule according to which different examples (images) are generated.
From examples, rules can be inferred. For example: with a sufficient dataset of images and their names, a DALLE-3 model can be trained on it.
In computer science, there is a concept called Kolmogorov complexity of data. It is (roughly) defined as the length of the shortest program capable of producing that data.
Some data are simple and can be compressed easily; some are complex and harder to compress. In a sense, the task of machine learning is to find a program of a given size that serves as a “compression” of the dataset.
In the real world, although knowing the underlying rule is often very useful, sometimes it is more practical to use a giant look-up table (GLUT) of examples. Sometimes you need to memorize the material instead of trying to “understand” it.
Sometimes there are examples that are more complex than the rule that generated them. For example, in the interval [0;1] (which is quite easy to describe, the rule being: all numbers are not greater than 1 and not less than 0), there exists a number containing all the works of Shakespeare (which definitely cannot be compressed to a description comparable to that of the interval [0;1]).
Or, сonsider the program that outputs every natural number from 1 to 101020 (which is very short, because the Kolmogorov complexity of 101020 is low) will at some point produce a binary encoding of LOTR. In that case, the complexity lies in the starting index, the map for finding the needle in the haystack is as valuable (and as complex) as the needle itself.
Properties follow from rules. It is not necessary to know about every example of a rule in order to have some information about all of them. Moreover, all examples together can have less information (or Kolmogorov complexity) than sum of individual Kolmogorov complexities (as in example above).