tbalsam

11 ms·

61.
by tbalsam 3y ago
The reasoning in the post is not backwards reasoning, it is literally the information theory of compression, described in human-friendly terms! But I must add the caveat that the original post is really only saying that it's rather tri
62.
by tbalsam 3y ago
Technically, under a compressive regime in the limit, LLMs necessarily need to encode an implicit world model, as the text input into is sampled from said world model. Variety, amount of training samples, capacity of the model, etc come int
63.
by tbalsam 3y ago
No, Agile is about tight development loops. When weaponized/used by large corporations it often times turns into a Stay-Puft man of sorts, but as someone who _hates_ processes normally, I actually kinda like it and Kanban when done wel
64.
by tbalsam 3y ago
Nightshade is intended to make minimal changes so I think that might be one argument though a difficult sell to some. The original post is mainly about the current legality, I'm torn as to what the moral balance is, but poisoning data
65.
by tbalsam 3y ago
I would recommend at least on the legal side of things looking into the Duty of Care history, it's not necessarily intuitive though I think it makes sense in the end. As another person said, oftentimes it comes down to intent (though n
66.
by tbalsam 3y ago
(I understand that this is not a popular point, but I really want to emphasize that I am talking about what is _currently legal_ right now, not at all about the ethicality of large companies using people's data. The latter is a much ha
67.
by tbalsam 3y ago
Currently it's not legally defined what is stealing and what is fair use with respect to this, which is of course why k feel it is such a strong issue, however, poisoning data and intentionally masking it to hide said poisoning is rath
68.
by tbalsam 3y ago
Duty of Care, additionally from an alternative perspective there are a number of laws covering the transmission of data that is intended to cause harm to another system.
69.
by tbalsam 3y ago
I think it is going to be a hard problem, at least. <3 :'))))
70.
by tbalsam 3y ago
Yes, this is one other slice of the pie that gives a more accurate picture of what it is looking at. The struggles of autism are significant, and while people on the spectrum make excellent researchers/engineers and such, many absolute
71.
by tbalsam 3y ago
What. Most people are high functioning ASD, it's a pareto distribution like almost every neurodivergent condition. Also, a shocking amount of competent researchers are autistic. As to the cause issue, yes, many potential causes, classi
72.
by tbalsam 3y ago
Hi! I'm rather autistic. This is a straightforward consequence of the bias-variance tradeoff, autistic brains tend to get too much information which is mathematically indistinguishable from noise. This is why many of the repetitive pat
73.
by tbalsam 3y ago
DMSO is terrifying, dude, it makes skin so permeable as I understand that supposedly even live bacteria outside of the skin can be transferred inside the skin. It's something I'm shocked seeing people casually use, similar to pipe
74.
by tbalsam 3y ago
I wonder if this a similar process the human brain goes through im certain 'normal', but still very extreme circumstances like excluding certain out-group members. I don't advocate for it, but I think we try to look away at w
75.
by tbalsam 3y ago
Ah, thanks, very coolio! Great to see this as a feature, haven't quite seen that mechanic well and I think it's a really great way of implementing it! :D :)
76.
by tbalsam 3y ago
Yes, though thankfully ondnanestron has a shockingly good safety profile for a serotonergic antiemetic. That said, reducing that external complexity would be nice. CBD actual has some mild antiemesis properties as well and can be as immedia
77.
by tbalsam 3y ago
Yes I think this is it right here. Like, at a micro dose (or slightly above) level it maybe makes sense, but the raw characteristics of how psychedelics work will still be there. Maybe there's some way, who knows, but I don't thin
78.
by tbalsam 3y ago
Please read in context with the rest of the post! The idea is that it's very hard to raise the entropy of the brain without psychomometic effects as best as I understand, due to some fundamental informational qualities of those process
79.
by tbalsam 3y ago
Oh, is this a pinned post? First time I think I've ever seen a post with so few votes do this before! :D
80.
by tbalsam 3y ago
While I'm not the biggest fan of companies trying to land grab the psychedelic craze, there are things that would be nice to have, like psychedelics that don't activate 5-HT2B (associated strongly with aortic hypertrophy, i.e., yo
81.
by tbalsam 3y ago
I read your post, and this is part of the phrasing that I am urging caution against, you are expecting a certain kind of connectivity of information within a non-linear model and attributing a deception hypothesis to it. Confabulation happe
82.
by tbalsam 3y ago
Right, language quickly moves beyond grammar within a certain point in the compressive regime of a model I.M.P.E., at leasts. <3 :'))))
83.
by tbalsam 3y ago
Ultimately unfortunately you go pretty far beyond the concepts of the English language pretty quickly here, I recommend the original Shannon paper followed by Varley's 2023 topic survey to get up to speed on it. Having English majors m
84.
by tbalsam 3y ago
Hi gwern. The other issue at hand is that inherently information is not always two-way between distributions, so having an implicit bias towards reversal actually can cause quite a few issues as well (though I'm unfortunately still in
85.
by tbalsam 3y ago
(moving this comment to this thread): 1. Models learn based on their data. Yes, data can be poisoned, but there's not much way around this. 2. Non-linear models, unlike linear models, have increasingly 'many' 'surfaces&#
86.
by tbalsam 3y ago
(comment moved to other thread).
87.
by tbalsam 3y ago
Apologies, I didn't realize it didn't allow for an invite, it's an open server: https://discord.com/invite/tfs2
88.
by tbalsam 3y ago
Team Fortress 2: Source 2 is down as well. https://discord.com/channels/902532397877297242/902540593606... To my knowledge, this is the first time Valve has done anything like this. They are legendary for letting
89.
by tbalsam 3y ago
Even in the days where this was standard, that is not the case entirely. There is a whole other world between "released code" and "getting the results as seen in the paper". Unfortunately. The reproducibility crisis is v
90.
by tbalsam 3y ago
Doesn't Phi-2 have testing data contamination, hence why it's performing well on these benchmarks? Most professionals in the field that I'm near would not touch that model with a 10 foot pole. We desperately need better valid

More ›