So you’re basically saying that, in your opinion, tensor operations are too simple of a building block for understanding to ever appear out of them as an emergent behavior? Do you feel that way about every mathematical and logical operation that a high school student can perform? That they can’t ever in whatever combination create a system complex enough for understanding to emerge?
I don’t think that anyone would argue that the general public can even solve a mathematical matrix, much less that they can only comprehend a stool based on going down a row in a matrix to get the mathematical similarity between a stool, a chair, a bench, a floor, and a cat.
LLMs rely on billions of precise calculations and yet they perform poorly when tasked with calculating numbers. Just because we don’t calculate anything consciously to get a meaning of a word doesn’t mean that no calculations are actually done as part of our thinking process.
What’s your definition of “the actual meaning of the concept represented by a word”? How would you differentiate a system that truly understands the meaning of a word vs a system that merely mimics this understanding?
technology fundamentally operates by probabilisticly stringing together the next most likely word to appear in the sentence based on the frequency said words appeared in the training data
What you’re describing is Markov chain, not an LLM.
So long as a model has no regard for the actual you know, meaning of the word
It does, that’s like the entire point of word embeddings.
Github link for convenience
https://github.com/Anuken/Mindustry
GPT3 is pretty bad at it compared to alternatives (although it’s hard to compete with calculators on that field), but if it was just repeating after the training dataset it would be way worse. From the study I’ve linked in my other comment (https://arxiv.org/pdf/2005.14165.pdf):
On addition and subtraction, GPT-3 displays strong proficiency when the number of digits is small, achieving 100% accuracy on 2 digit addition, 98.9% at 2 digit subtraction, 80.2% at 3 digit addition, and 94.2% at 3-digit subtraction. Performance decreases as the number of digits increases, but GPT-3 still achieves 25-26% accuracy on four digit operations and 9-10% accuracy on five digit operations, suggesting at least some capacity to generalize to larger numbers of digits.
To spot-check whether the model is simply memorizing specific arithmetic problems, we took the 3-digit arithmetic problems in our test set and searched for them in our training data in both the forms “<NUM1> + <NUM2> =” and “<NUM1> plus <NUM2>”. Out of 2,000 addition problems we found only 17 matches (0.8%) and out of 2,000 subtraction problems we found only 2 matches (0.1%), suggesting that only a trivial fraction of the correct answers could have been memorized. In addition, inspection of incorrect answers reveals that the model often makes mistakes such as not carrying a “1”, suggesting it is actually attempting to perform the relevant computation rather than memorizing a table.
In my comment I’ve been referencing https://arxiv.org/pdf/2005.14165.pdf, specifically section 3.9.1 where they summarize results of the arithmetic tasks.
That’s not entirely true.
LLMs are trained to predict next word given context, yes. But in order to do that, they develop internal model that minimizes error across wide range of contexts - and emergent feature of this process is that the model DOES perform more than pure compression of the training data.
For example, GPT-3 is able to calculate addition and subtraction problems that didn’t appear in the training dataset. This would suggest that the model learned how to perform addition and subtraction, likely because it was easier or more efficient than storing all of the examples from the training data separately.
This is a simple to measure example, but it’s enough to suggests that LLMs are able to extrapolate from the training data and perform more than just stitch relevant parts of the dataset together.
They can replace them going forward. A major issue is that many governments (and likely other malicious actors) have been hoarding encrypted communication in hopes of accessing it once sufficiently big quantum computer emerges.
Have you tried Cosmoteer? It’s a pretty satisfying shipbuilder with resource and crew management, trading, and quests. Similar vibe to Reassembly.