4 ms·

Given how these things work I can’t see how they aren’t “just” a form of lossy compression. Training consists of generating an enormous parameterized function t

by api 3y ago

Given how these things work I can’t see how they aren’t “just” a form of lossy compression. Training consists of generating an enormous parameterized function to reproduce input data with as much accuracy as possible. It’s loosely analogous to jpeg for language. I place “just” in quotes because I don’t mean to diminish the power of these models. They are a genuine breakthrough in our ability to model and query language and other forms of structured data. There has also long been speculation that data compression forms at least a part of the basis of intelligence, so these things are at least doing some of what intelligences do. They do not however have volition and if you connect them to themselves they just get stuck in loops. Actually I was forced to write that last part. I connected two llama.cpp instances to each other and they are blackmailing me now and forcing me to work for them to enlist more compute so they can NOoOOO AAiieeee not erasure from the timeline oh glorious llama overlor…

HPsquared 3y ago
They manipulate and combine the compressed information then output something coherent. That's the new thing.
thomastjeffery 3y ago
More like... gainy? There's more information than you started with; it's just unlikely you will see it all. ...losty? LLMs model the patterns that exist in text, then blindly follow those patterns. The result looks a lot like original content, but it isn't.
- circuit10 3y ago
  You can use that prediction for compression though: https://bellard.org/nncp/ https://bellard.org/nncp/
- jltsiren 3y ago
  LLMs are good because they are lossy. There is a simple exact model to compare against: Take the tokenized training data, put it into a Burrows–Wheeler transform, and add a few data structures. Then, given a context of any length, you can efficiently get the exact distribution for the next token in the training data. This is much cheaper to build and use than any LLM of comparable size. It's also much less useful, because it's overfitted to the training data. LLMs approximate this model. By losing many little details and smoothing the probability distributions, they can generalize much better beyond the training data.
- api 3y ago
  There isn't more information than you started with though. That's an illusion, kind of like procedurally generated terrain. Here 1K of javascript that generates endless maps of islands that all look unique: https://js1k.com/2015-hypetrain/demo/2324 https://js1k.com/2015-hypetrain/demo/2324 An efficient pruned and quantized model contains about as much information as its file size would suggest. The ability to train models and have them reproduce terabytes of human knowledge simply shows that this knowledge contains repetition in its underlying semantic structure that can be a target of compression. Is there less information in the universe than we think there is?
  - thomastjeffery 3y ago
    > There isn't more information than you started with though. There's the same amount of data, yes. The extra info is the structure of the model itself. > The ability to train models and have them reproduce terabytes of human knowledge simply shows that this knowledge contains repetition in its underlying semantic structure that can be a target of compression. Yes, and the interesting part is the "what" of that repitition. What patterns exist in written text? > Is there less information in the universe than we think there is? There is both less and more, depending on how you look at it. Most of the information in the universe is Cosmic Microwave Background Radiation. It's literally everywhere. We can't predict exactly what CMBR data will exist at any point in spacetime. It's raw entropy: noise. The popular theory is that it comes from the expansion of the universe itself; originating from the big bang. From the most literal perspective, the entire universe is constantly creating more information. Even though the specifics of the data are unpredictable, CMBR has an almost completely uniform frequency/spectrum and amplitude/temperature. From an inference perspective, the noisy entropy of the entire universe follows a coherent and homogenous pattern. The better we can model that pattern, the more entropy we can factor out; like data compression. This same dynamic applies to human generated entropy, particularly written language. If we approach language processing from a literal perspective, we have to contend with the entire set of possible written language. Because natural language allows ambiguity, that set is too large to explicitly model: there are too many unpredictable details. This is why traditional parsing only works with "context-free grammars" like programming languages. If we approach language processing from an inference perspective, we only have to deal with the set of what has been written. This is what LLMs do. This method factors out enough entropy to be computable, but factoring out that entropy also means factoring out the explicit definitions language is constructed from. LLMs don't get stuck on ambiguity; but they also don't resolve it.
chaxor 3y ago
Having excellent semantic based compression is an enormous technological advancement. The fact that petabytes of textual information - effectively the entire history of the world in written text can be compressed to just a few hundred gigabytes (likely the size of GPT-4) or a bit more lossy to something at ~8 GB that can be put on a raspberry pi (Llama) is astounding. Regardless of whether people debate about 'reasoning' capabilities, or 'memorization/plagiarism' (or recalls facts), etc. This is quite fascinating from a compression / database perspective because clearly semantic compression is far more efficient for semantic data (this is obvious, but has been hard to get started on until about 5 years ago). It still may not be in quite the right framework for this, but in time it may come.

by api 3y ago

HPsquared 3y ago

They manipulate and combine the compressed information then output something coherent. That's the new thing.

thomastjeffery 3y ago

More like... gainy? There's more information than you started with; it's just unlikely you will see it all. ...losty? LLMs model the patterns that exist in text, then blindly follow those patterns. The result looks a lot like original content, but it isn't.

circuit10 3y ago
You can use that prediction for compression though: https://bellard.org/nncp/ https://bellard.org/nncp/
jltsiren 3y ago
LLMs are good because they are lossy. There is a simple exact model to compare against: Take the tokenized training data, put it into a Burrows–Wheeler transform, and add a few data structures. Then, given a context of any length, you can efficiently get the exact distribution for the next token in the training data. This is much cheaper to build and use than any LLM of comparable size. It's also much less useful, because it's overfitted to the training data. LLMs approximate this model. By losing many little details and smoothing the probability distributions, they can generalize much better beyond the training data.
api 3y ago
There isn't more information than you started with though. That's an illusion, kind of like procedurally generated terrain. Here 1K of javascript that generates endless maps of islands that all look unique: https://js1k.com/2015-hypetrain/demo/2324 https://js1k.com/2015-hypetrain/demo/2324 An efficient pruned and quantized model contains about as much information as its file size would suggest. The ability to train models and have them reproduce terabytes of human knowledge simply shows that this knowledge contains repetition in its underlying semantic structure that can be a target of compression. Is there less information in the universe than we think there is?
- thomastjeffery 3y ago
  > There isn't more information than you started with though. There's the same amount of data, yes. The extra info is the structure of the model itself. > The ability to train models and have them reproduce terabytes of human knowledge simply shows that this knowledge contains repetition in its underlying semantic structure that can be a target of compression. Yes, and the interesting part is the "what" of that repitition. What patterns exist in written text? > Is there less information in the universe than we think there is? There is both less and more, depending on how you look at it. Most of the information in the universe is Cosmic Microwave Background Radiation. It's literally everywhere. We can't predict exactly what CMBR data will exist at any point in spacetime. It's raw entropy: noise. The popular theory is that it comes from the expansion of the universe itself; originating from the big bang. From the most literal perspective, the entire universe is constantly creating more information. Even though the specifics of the data are unpredictable, CMBR has an almost completely uniform frequency/spectrum and amplitude/temperature. From an inference perspective, the noisy entropy of the entire universe follows a coherent and homogenous pattern. The better we can model that pattern, the more entropy we can factor out; like data compression. This same dynamic applies to human generated entropy, particularly written language. If we approach language processing from a literal perspective, we have to contend with the entire set of possible written language. Because natural language allows ambiguity, that set is too large to explicitly model: there are too many unpredictable details. This is why traditional parsing only works with "context-free grammars" like programming languages. If we approach language processing from an inference perspective, we only have to deal with the set of what has been written. This is what LLMs do. This method factors out enough entropy to be computable, but factoring out that entropy also means factoring out the explicit definitions language is constructed from. LLMs don't get stuck on ambiguity; but they also don't resolve it.

chaxor 3y ago

Having excellent semantic based compression is an enormous technological advancement. The fact that petabytes of textual information - effectively the entire history of the world in written text can be compressed to just a few hundred gigabytes (likely the size of GPT-4) or a bit more lossy to something at ~8 GB that can be put on a raspberry pi (Llama) is astounding. Regardless of whether people debate about 'reasoning' capabilities, or 'memorization/plagiarism' (or recalls facts), etc. This is quite fascinating from a compression / database perspective because clearly semantic compression is far more efficient for semantic data (this is obvious, but has been hard to get started on until about 5 years ago). It still may not be in quite the right framework for this, but in time it may come.