aezart

8 ms·

91.
by aezart 3y ago
Not for a lack of trying. If you give it a small enough training set or a big enough neural network, it will directly memorize the whole thing. You have to intentionally make its brain too small to do that in order to force it to find pat
92.
by aezart 3y ago
The last time I tried Coke Zero, it tasted like Diet RC. I didn't like it at all. Diet Coke has huge differences in flavor between sources and packaging options. It's universally bad out of a can or bottle, and the stuff from soda
93.
by aezart 3y ago
That's quite possible, I'm not super into the modding scene. I guess my default assumption is that mods should be running through some sort of scripting interface that doesn't have access to things like the filesystem.
94.
by aezart 3y ago
I foresee this being a big problem for the Godot game engine as it gains popularity. There's absolutely no sandboxing or "safe mode" or anything when loading extra resource packs, and any resource or scene file can have an em
95.
by aezart 3y ago
Rapid degradation, like recording multiple generations of VHS tapes. The LLMs make the internet dumber, the LLMs get dumber by learning from it. Rinse and repeat.
96.
by aezart 3y ago
Yeah, a third output would be nice to have. For example, grep could use stdout to print the actual matches it finds, stderr for any files it was unable to open, and a third stream to show the progress of the search.
97.
by aezart 3y ago
I think they're ugly as hell and in poor taste but I also think people should have the right to do what they want with their homes as long as it's not dangerous. Although since they're on a sidewalk it's a problem if the
98.
by aezart 3y ago
"US Bank deposits down $427B in Q1 2023" would make the most sense. I also read "lost" as "lost track of", as in "we received these deposits but then don't know where they went".
99.
by aezart 3y ago
I've had bad luck with third-party toner cartridges on my Brother printer. Eventually it just stop detecting them at all.
100.
by aezart 3y ago
Typically it takes 2 years instead of the 4 a bachelor's degree would take
101.
by aezart 3y ago
Yeah, from the trailers and footage I've seen on Twitter this doesn't seem like the game for me. I want a dungeon crawler, not a physics sandbox. Also I want Link to be left handed again.
102.
by aezart 3y ago
Oh no, another one? We already have a buffelgrass problem.
103.
by aezart 3y ago
A time waster's not always a bad thing; I spent lots of time browsing it back when I was a student intern because it seemed on topic" enough for me to not feel guilty.
104.
by aezart 3y ago
It's all based on attention, right? If the model winds up giving a low attention value to the delimiters, they won't help at all.
105.
by aezart 3y ago
I didn't get credit for the first few assignments in my first CS class because the automated grading tool couldn't find my code. Finally I figured out that the tool only worked if the code was in the default Java package, which I
106.
by aezart 3y ago
The funniest outcome of all this would be a ruling that the entire copilot model is GPL.
107.
by aezart 3y ago
I wound up just getting a signed cert for my home webserver. It wasn't too painful.
108.
by aezart 3y ago
That's what github enterprise is for
109.
by aezart 3y ago
One thing I'm not clear about is how the LLM decides on the correct tokenization to use. Looking at GPT2's token vocabulary, it has the word "the" as token number 1169, but it also has "t", "h", and &
110.
by aezart 3y ago
I can already run a language model on my GPU using minillm or text-generation-webui, and on my CPU using llama.cpp. What makes MLC-LLM better?
111.
by aezart 3y ago
Yeah, you have to exactly match electric output to demand or things get very bad. Until we have a few more breakthroughs in battery technology, we're going to be stuck with fossil fuels to make up the gap for variable demand. Nuclear c
112.
by aezart 3y ago
The point is that the American units are now defined relative to metric units. As recently as the 1800s, the yard was defined as: "the Straight Line or Distance between the Centres of the Two Points in the Gold Studs of the Straight Br
113.
by aezart 3y ago
LLaMA claims to be GPT-4 when I ask it what it is, so I don't necessarily trust that response.
114.
by aezart 3y ago
From my limited understanding of how LLMs work, I believe this behavior is enabled by embeddings. The model maps its ~50,000 token vocabulary into a lower-dimensional vector space. Each dimension in the vector adds some sort of meaning (o
115.
by aezart 3y ago
True but not very useful. That's like writing an "isPrime" function with a switch statement for every possible input number, rather than coming up with a general algorithm. It can't give you any answer you don't alr
116.
by aezart 4y ago
I agree. A stopped clock is right twice a day, and that's equally true with these LLMs.
117.
by aezart 4y ago
LLM output cannot be higher quality than the input (prompt + training data). The best possible outcome for an LLM is that the output is a correct continuation of the prompt. The output will usually be a less-than-perfect continuation. With
118.
by aezart 4y ago
Obviously, the only solution is to revert to an oral tradition. Stop writing, and instead pass knowledge from generation to generation in the form of poems and stories told by word of mouth.
119.
by aezart 4y ago
I'm very worried that machine learning will lead to stagnation in general, in programming and in art. People will accept the machine output and call it a day instead of making something new.
120.
by aezart 4y ago
This is my worry as well. Google results are already full of content farm/AI SEO junk, and I don't understand people who see that and think "let's do that, but more". There's already more good content than I

More ›