Why not apply same concept every time a word is split into more than one token?
Basically if a word contain a Prefix, suffix or root word. We could have a token position relative to the start of the word in the embedding.
Transformers Can Do Arithmetic with the Right Embeddings
131–140 of 215 posts
Re: Transformers Can Do Arithmetic with the Right Embeddings
#132Earlier quoted context omitted.
> humans follow a deductive process to answer questions which they haven't encountered yet nope. most humans fall in various traps such as pattern recognition, confirmation bias, and many others instead of relying on deductive analysis. Even scientists fail at being rigorous.
Of course there are cases like this, nobody is perfect. But we are talking about mathematics here, not everyday subconscious decision making. I agree that 99% of daily life is trivial pattern recognition. That's not what distinguishes humans though is it? Because animals, down to single celled organisms do just fine without higher order mental capabilities. But we are talking about reasoning here - and specifically a…
I disagree that daily life is "trivial pattern recognition".
Just our visual object recognition is immensely powerful and far beyond and current AI. A simple task like walking to the fridge requires a ton of pattern recognition and spatial reasoning. Recognizing people's moods/predicting behaviors is also incredibly involved imo.
Ive said this many times but perhaps we should focus on achieving dog level intelligence first before we start worrying about human level AGI.
Re: Transformers Can Do Arithmetic with the Right Embeddings
#133Why not apply same concept every time a word is split into more than one token? Basically if a word contain a Prefix, suffix or root word. We could have a token position relative to the start of the word in the embedding.
It seems it has been done before:
"Syntax-Aware Transformer Models for Neural Machine Translation" by Yang et al. (2019). This model enhances the transformer architecture with syntax-aware attention mechanisms that consider dependency parse trees.
Context-Aware Neural Machine Translation Learns Anaphora Resolution" by Bawden et al. (2018). This paper explores integrating context and syntax into neural machine translation models.
Re: Transformers Can Do Arithmetic with the Right Embeddings
#134Earlier quoted context omitted.
> However, by itself, it seems like this capability wouldn't be very useful, commercially for example. Quite the opposite, it's the holy grail of all AI. Consider various work that isn't (and can't) be done by computers/robots/etc right now. The intelligence constraint is universally, a required amount of problem solving. Even the "low skill" labour requires it. And to perform such problem solving, you need advanced…
Let's be a little more concrete: do you think FormalGeo [1] is a big deal? I think it's very cool but ultimately not useful in and of itself. It's only useful insofar as it shows AI capabilities advancing in general. Let's suppose we had an AI that works roughly like [1] but for the kind of mathematics done in Lean's Mathlib, and that was on par or better than humans working on it. Would that AI by itself be commerci…
> I just can't think of much besides a major advance in the formal software verification niche, which still almost nobody would use...
The reason is slightly different here.
What's so desirable here is an AI system with such general intelligence that it is capable of such mathematics by itself as a consequence. Not because the mathematics is so useful, but because the required reasoning capabilities are at such a level that, we could speak of an artificial intelligence that is meaningfully "general" about any problem.
It's a decent approximation of "able to solve any problem" that we can still reasonably test.
> Let's be a little more concrete: do you think FormalGeo [1] is a big deal?
It looks to be an interesting approach in modelling mathematics, and their use of machine learning is an interesting novelty that may pave the way to more useful general mathematics systems, but I can't find much about how these systems might interop with current/'generative' AI systems.
And that last bit is one of the big roadblocks for current AI. They're very weak at reasoning, but we can't directly interop to (e.g.) LLMs, so we can't compensate for that weakness.
Re: Transformers Can Do Arithmetic with the Right Embeddings
#135Earlier quoted context omitted.
how do you argue that these models are not able to reason? deductive reasoning is just drawing specific conclusion from general patterns. something I would argue this models can do (of course not always and are still pretty bad in most cases) the point i’m trying to make is that sometimes reasoning is overrated and put on the top of the cognitive ladder, sometimes I have seen it compared to self-awareness or stuff li…
>> deductive reasoning is just drawing specific conclusion from general patterns. This is according to whom, please?
> Deductive reasoning is the process of drawing valid inferences. An inference is valid if its conclusion follows logically from its premises, meaning that it is impossible for the premises to be true and the conclusion to be false.
https://en.wikipedia.org/wiki/Deductive_reasoning>
Re: Transformers Can Do Arithmetic with the Right Embeddings
#136Earlier quoted context omitted.
> This is basically free to add, and there's no reason it shouldn't be made part of standard tokenization. This is muchhhhh different from how tokenization works today. Adding tokens to the vocabulary is free, everything outside that (i.e. string -> tokens) is going to be a major pain in the ass. Doable but annoying and error prone
Doesn’t seem as complicated as, say, coding a lexer for C. And why shouldn’t tokenisation use lexers or an equivalent?
Good old software development. :( Recent case studies:
- llama.cpp wasn't tokenizing properly, and it came to a head with llama3. Essentially every local model before May 2024 is soft-deprecated, new ones have to indicate the proper tokenizer, and that currently only covers a small subset of popular models
- I recently had to review 41 Phi-3 and Llama 3 models, only 3 had the right tokenizer set
Not saying it's impossible, and we definitely should, and I bet it 100% happens, but...*shudders*
Re: Transformers Can Do Arithmetic with the Right Embeddings
#137Earlier quoted context omitted.
Of course there are cases like this, nobody is perfect. But we are talking about mathematics here, not everyday subconscious decision making. I agree that 99% of daily life is trivial pattern recognition. That's not what distinguishes humans though is it? Because animals, down to single celled organisms do just fine without higher order mental capabilities. But we are talking about reasoning here - and specifically a…
I disagree that daily life is "trivial pattern recognition". Just our visual object recognition is immensely powerful and far beyond and current AI. A simple task like walking to the fridge requires a ton of pattern recognition and spatial reasoning. Recognizing people's moods/predicting behaviors is also incredibly involved imo. Ive said this many times but perhaps we should focus on achieving dog level intelligence…
Oh I'm very much with you. In fact I get irked by people here breathlessly parroting that human level AGI is upon us any day now. I'd be impressed if an AI had mouse level capabilities any time soon. I think the current models are very impressive, but they are parlor tricks compared to what a true AGI should be capable of.
Re: Transformers Can Do Arithmetic with the Right Embeddings
#138Earlier quoted context omitted.
I disagree that daily life is "trivial pattern recognition". Just our visual object recognition is immensely powerful and far beyond and current AI. A simple task like walking to the fridge requires a ton of pattern recognition and spatial reasoning. Recognizing people's moods/predicting behaviors is also incredibly involved imo. Ive said this many times but perhaps we should focus on achieving dog level intelligence…
Oh I'm very much with you. In fact I get irked by people here breathlessly parroting that human level AGI is upon us any day now. I'd be impressed if an AI had mouse level capabilities any time soon. I think the current models are very impressive, but they are parlor tricks compared to what a true AGI should be capable of.
>if an AI had mouse level capabilities any time soon
That's why nobody has gotten any traction selling access to AIs for $20 a month whereas selling access to mouse labor is such a thriving business.
Re: Transformers Can Do Arithmetic with the Right Embeddings
#139I think the problem here is that 'understanding' is not the same as curve fitting. If all one is doing is giving a model lots of data and fitting curves it's not really 'understanding' but brute forcing it's way (with gradient descent) and then storing the weights and finally approximate the solution when a query is passed in. This is not the same as understanding. Human intelligence can operate deterministically as…
This seems like its confusing how we conceptualize the training/learning process with what the system is actually doing. We conceptualize tuning parameters as curve fitting, and we conceptualize predicting the next token as maximizing probability. But that doesn't mean there is anything like curve fitting or probability maxxing happening as the system's parameters converge.
The core feature of curve fitting is learning explicit examples and then interpolating (in an uninformative manner) between unlearned examples. But there's no reason to think this completely describes what the system is doing, in the sense that there are no more informative descriptions of its behavior. Take an example that LLMs are surprisingly good at, creating poetry given arbitrary constraints. Imagine the ratio of the poems it has seen during its training over the number of unique poems it could create in principle. This number would be vanishingly small. Interpolating between two strings representing well-formed poems in an uninformative manner (i.e. some finite polynomial) will not generate well-formed poems. The only way you could move between two examples of well-formed poems while staying on the manifold of well-formed poems is if you captured all relevant features of the manifold. But I fail to see a difference between capturing all relevant features of the poetry-manifold and understanding poetry.
What LLMs do can be described as curve fitting in only the most uninformative description possible. What they do is discover features of the structures referred to by the training text and competently deploy these features in predicting the next token. A human that could do this would be consider to understand said structure.
Re: Transformers Can Do Arithmetic with the Right Embeddings
#140I went through the paper and thought immediately about how did they implement it; I missed they published their code as well. Here is the link for everyone who skimmed past it: https://github.com/mcleish7/arithmetic/tree/main