genrilz

10 ms·

31.
by genrilz 2y ago
I think what you are saying here is that because it has no "concept" (I'll assume that means internal model) of truth, then there is no possible way of improving the truthiness of an LLMs outputs. However, we do know that LLM
32.
by genrilz 2y ago
I'm not sure why you think hallucinations can't be "fixed". If we define hallucinations as falsehoods introduced between the training data and LLM output, then it seems obvious that the hallucination rate could at least
33.
by genrilz 2y ago
Overall, excellent points! I would like to add on to that though. RLHF actually does effectively have one LLM educating another. Specifically, human trainer's time is valuable, so they train an AI to express the same opinion about some
34.
by genrilz 2y ago
I mean, sure. OP implied that "capital of Connecticut is Moscow" is the sort of thing that a human "model of truthfulness" would encode. I'm pointing out that the human model of that particular fact isn't inher
35.
by genrilz 2y ago
We kinda do have LLMs in a swarm configuration though. Currently LLMs training data, which includes all of the non RAG facts they know, come from the swarm that is humans. As LLM outputs seep into the internet, older generations effectively
36.
by genrilz 2y ago
There is some model of truthfulness encoded in our heads, and we don't draw all of that from our direct experience. For instance, I have never been to Connecticut or Moscow, but I still think that it is false that the capital of Connec
37.
by genrilz 2y ago
I acknowledge that all lies are going to make things more difficult in the long run while making things easier in the short run. I would like to point out that sometimes things need to be easier in the short run though. In code terms this i
38.
by genrilz 2y ago
I think I spelled it out in the post you are referencing, but I see a white lie as the equivalent of "not saying something because it is rude", except for when you are forced to say something and not doing so would be meaningful.
39.
by genrilz 2y ago
I agree that in professional settings, it's usually best for the efficiency of the company to be direct about everything. I also generally prefer that people provide accurate feedback about how I am doing in interpersonal relationships
40.
by genrilz 2y ago
I believe you when you say that white lies violate your trust. I'm saying that that experience isn't universal though. Different people define violations of their trust differently. Thus my suggestion at the end to make sure that
41.
by genrilz 2y ago
I think what GP is saying is that white lies are not necessarily betrayal. Betrayal is when someone violates your trust somehow. [0] You can see some examples of lying in the wiktionary definition I linked. (see the last two definitions) Ho
42.
by genrilz 2y ago
I think we are speaking past each other here. I'm talking in consequentialist terms and you are (I think, feel free to correct me if I am wrong) talking in virtue ethics terms. I'm assuming you aren't familiar with these term
43.
by genrilz 2y ago
Aren't the broad principles how we make the finer grained judgements in a fair and consistent manner though? Judges are at least supposed to use the outcomes of previous cases and the letter of the law to do this. The rest of us need s
44.
by genrilz 2y ago
On the utilitarian front, I would prioritize my mother, but it would be because I care about her, not because I would be acting morally. I accept that I can't ever be a perfect utilitarian. Perfect is the enemy of good though. I don&#x
45.
by genrilz 2y ago
Sorry about the delay. I got sidetracked. That... is actually a pretty interesting argument. I have to admit that if an objective morality existed floating in the Aether, there would be no way to logically prove or disprove that one's
46.
by genrilz 2y ago
While I personally also consider my relationships to be more important than my earnings, I am still going to argue that it's subjective. Case in point, both you and I disagree with Altman about what success means. We are all human bein
47.
by genrilz 2y ago
EDIT: I got arguments mixed up. It's master-lincoln who was making the health arguments, not you. You were talking about not holding women to unreasonable beauty standards. I mostly agree, although how to actually stop doing that is ve
48.
by genrilz 2y ago
I feel like we might have been reading the original arguments in different ways, so let me summarize how I see this thread: I interpreted ndndjdjdn's comment as sarcasm. (due to the use of the phrase "next up") That is, I thi
49.
by genrilz 2y ago
This is actually probably more fair of a comparison then you'd think. Daily showers are bad for health your health[0], but I absolutely do them because my body produces a lot of oil and odor. This is a cosmetic reason similar to make-u
50.
by genrilz 2y ago
I feel like that is a pretty high bar to call "reasoning". I would like to think I am capable of reasoning, and yet I would not be able to write out by hand a binary file to be loaded on to the controller without using a compiler
51.
by genrilz 2y ago
I think it is annoying to have to copy past the link in order to get around the essay too. I think its fine to say so. Its the people saying that HN isn't actually like how the essay represents HN to be that my original post and previo
52.
by genrilz 2y ago
The comment is not something someone would say to a person they intend to treat as a peer. I personally wouldn't post a comment like that unless I thought the person I was talking about was a danger to other people, regardless of how I
53.
by genrilz 2y ago
I do have some memories of not particularly trans positive comments in the comments section on earlier blogposts about their opengl/vulkan drivers, and also references to kiwi farms. I don't see such things commonly on most hacker
54.
by genrilz 2y ago
For this model, it seems like the point is that it uses way less parameters than at least the large Llama model while having near identical performance. Given how large these models are getting, this is an important thing to do before makin
55.
by genrilz 2y ago
It seems like you are getting down-voted, but I think you are probably right. One concern I have is that publicly announcing ways to break the law might result in the government cracking down on those particular ways. You could make a case
56.
by genrilz 2y ago
I am certainly glad that you managed to escape. I don't think our society's response to the welfare cliff should be to tell people to break the law though. Surely we should redesign the law to lift everyone up instead instead?
57.
by genrilz 2y ago
Obviously, there are some cases where it makes sense to use an unsafe block. However, I think there might be fewer cases then people might think. As an example, both the popular generic self-referencing crates ouroboros and self_cell have h
58.
by genrilz 2y ago
I would like to note that he has created/supported a variety of political causes in the past. His current cause does seem to be pointing to restricting kids internet access, but I can't really figure out what he wants. One of thes
59.
by genrilz 2y ago
I think I actually failed to respond properly in this comment. You were talking about smartphone use while the article was talking about social media use. I think Haidt and implicitly both of us were conflating the two, but they are separat
60.
by genrilz 2y ago
I've done literature reviews after seeing posts from Haidt come up here and elsewhere. If you go to semanticscholar.org for instance, type in "social media wellbeing meta-analysis", you will get something like [0]. (found a f

More ›