Jack000

11 ms·

61.
by Jack000 4y ago
The vae used in stable diffusion is not ideal for compression. I think it would be better to use the vector-quantized variant (by the same authors of latent diffusion) instead of the KL variant, then store the indexes for each quantized vec
62.
by Jack000 4y ago
the grid is meant to show the effects of different settings on the final image. Each row is a different seed, and each column is a different guidance scale - from 0, 50, 100 -> 500 in some cases the classifier guidance isn't strong
63.
Show HN: Classifier Guided Stable Diffusion(github.com)
1 points by Jack000 4y ago|2 comments
64.
by Jack000 4y ago
they should set up a fractional reserve system, where you deposit your valuable cards at a card bank and get a "card voucher" you can then use to play actual games with. The card bank could then lend out the cards to other players
65.
by Jack000 4y ago
ML research isn't focused on art but image generation. Art is simply what people are using these models for. supervised training of large models requires a large amount of data. The only realistic source for this data is the internet,
66.
by Jack000 4y ago
This technology can be applied beyond images, to virtually any data modality. - Video + audio generation (or automatic tweening for animation) means a single animator can create an entire movie or animated series. - 3d models and textures +
67.
by Jack000 4y ago
AI: remake seasons 5-8 of game of thrones, based on GRRM's final two books which he finally finished.
68.
by Jack000 4y ago
The scaling laws for LLMs depend heavily on the quality of data. For example, if you add an additional 100gb of data but it only contains the same repeating word, that will hurt the model. If you add 100gb of completely random words, that w
69.
by Jack000 4y ago
"Sentience" is a weird word because it's a such fuzzy concept, but I think if we define it properly there are testable answers to the question of machine sentience. I would define sentience as an awareness of oneself in relat
70.
by Jack000 4y ago
The scaling laws of large language models are very specific to language models and the way they're trained. The important thing that LLMs demonstrate is that transformers are capable of this kind of scale (where other approaches have n
71.
by Jack000 4y ago
I think it's fair to say that no one knows if AGI is even possible at this point, but RL+transformer is definitely the most promising approach imo. 10 years ago no one was talking about AGI, researchers didn't even call it "A
72.
by Jack000 4y ago
Yann LeCun's paper isn't really a traditional ML paper, it discusses ideas at a high level but doesn't mention anything about implementation. If you were to try to create an RL agent based on his ideas, with the perception en
73.
by Jack000 4y ago
Biology is not necessarily the only path to intelligence. Modern ML has diverged very far from biomimetic approaches at this point. The brain does have more "parameters" than ML models but this applies just as much to humans as it
74.
by Jack000 4y ago
Which part do you find objectionable - the lack of progress in previous years or the current/future potential of transformers in RL? I do work in ML but mostly applications instead of research.
75.
by Jack000 4y ago
Everything Deepmind published at this year's ICML would be a good start. Transformers (or rather the QKV attention mechanism) has taken over ML research at this point, it just scales and works in places it really shouldn't. Eg. yo
76.
by Jack000 4y ago
I'm very optimistic for near-term AGI (10 years or less). Even just a few years ago most in the field would have said that it's an "unknown unknown", we didn't have the theory or the models, there was no path forwar
77.
by Jack000 4y ago
Language model != Transformers. The scaling power of the transformer architecture allows LLMs to exist, but LMs are just one application of transformers. The architecture itself is extremely generic and is applicable to nearly every problem
78.
by Jack000 4y ago
There is a great short story by Greg Egan that deals with this issue called "Learning to be me". There's actually another short story that takes place in the same universe which kind of refutes this article imo.
79.
by Jack000 4y ago
Yeah that would work. I'm talking about LLMs, DALL-E and diffusion image generators specifically.
80.
by Jack000 4y ago
The reason that language models require large amounts of data is because they lack grounding. When humans write a sentence about.. let's say "fire", we can relate that word to visual, auditory and kinesthetic experiences buil
81.
by Jack000 4y ago
just put this sentence through a few grammar apps for comparison. grammarly: > And I would say even for a native speaker, not only a tool like this is for education and to improve one's confidence, but also demonstrated you care abo
82.
by Jack000 4y ago
these particular ML models are trained on human-generated data, which means the model is a reflection of that data. This doesn't mean that the model can't create something original though. There's a video on youtube where a l
83.
by Jack000 4y ago
I think there's some confusion over the terms graphic design, art, and image generation, which are somewhat overlapping but distinct fields. Dalle-2 doesn't really do graphic design, but a future version of it might.
84.
by Jack000 4y ago
hm.. it's spelled out pretty explicitly > Better Living Through Zoophilia. His long loony screed praising the utilitarian benefits of genetically engineering animals that would enjoy having sex with people. > Umm . . . you’re not
85.
by Jack000 4y ago
- the IP infringement is very real (see Nortel vs Huawei), but it's more the actions of individual companies than industrial policy. - Taiwan is technically the legitimate government of China, so it's really more appropriate to sa
86.
by Jack000 4y ago
For side projects intended to generate income, base it on market research. - Ensure there's demand. Go on google and type in your problem or solution, are there a lot of results and ads? If yes, there is demand. If the top results are
87.
by Jack000 4y ago
Wordle is a great name - it's short, only two syllables and relates directly to the product. A better example of a bad name might be Ycombinator. It's not immediately obvious what it means, and if you say it aloud it sounds like a
88.
by Jack000 4y ago
The idea that embodiment is necessary for intelligence is just another example of anthropocentrism. It's likely that we will soon have AGI that can solve the protein folding problem, play go at a superhuman level, and create arbitrary
89.
by Jack000 4y ago
I agree. You might as well replace the word "qualia" with "soul" and the argument stays exactly the same.
90.
by Jack000 4y ago
I think the main disconnect between large language models and what people expect from "AI" is the fact that LMs learn meaningful representations whereas most people associate intelligence with meaningful behaviors. All examples of

More ›