…Duh. 🤓
Someone put 69 to research and then to article. Nice trolling.
To understand what’s actually happening, Anthropic’s researchers developed a new technique, called circuit tracing, to track the decision-making processes inside a large language model step-by-step. They then applied it to their own Claude 3.5 Haiku LLM.
Anthropic says its approach was inspired by the brain scanning techniques used in neuroscience and can identify components of the model that are active at different times. In other words, it’s a little like a brain scanner spotting which parts of the brain are firing during a cognitive process.
This is why LLMs are so patchy at math. (Image credit: Anthropic)
Anthropic made lots of intriguing discoveries using this approach, not least of which is why LLMs are so terrible at basic mathematics. “Ask Claude to add 36 and 59 and the model will go through a series of odd steps, including first adding a selection of approximate values (add 40ish and 60ish, add 57ish and 36ish). Towards the end of its process, it comes up with the value 92ish. Meanwhile, another sequence of steps focuses on the last digits, 6 and 9, and determines that the answer must end in a 5. Putting that together with 92ish gives the correct answer of 95,” the MIT article explains.
But here’s the really funky bit. If you ask Claude how it got the correct answer of 95, it will apparently tell you, “I added the ones (6+9=15), carried the 1, then added the 10s (3+5+1=9), resulting in 95.” But that actually only reflects common answers in its training data as to how the sum might be completed, as opposed to what it actually did.
In other words, not only does the model use a very, very odd method to do the maths, you can’t trust its explanations as to what it has just done. That’s significant and shows that model outputs can not be relied upon when designing guardrails for AI. Their internal workings need to be understood, too.
Another very surprising outcome of the research is the discovery that these LLMs do not, as is widely assumed, operate by merely predicting the next word. By tracing how Claude generated rhyming couplets, Anthropic found that it chose the rhyming word at the end of verses first, then filled in the rest of the line.
“The planning thing in poems blew me away,” says Batson. “Instead of at the very last minute trying to make the rhyme make sense, it knows where it’s going.”
Anthropic discovered that their Claude LLM didn’t just predict the next word. (Image credit: Anthropic)
Anthropic also found, among other things, that Claude “sometimes thinks in a conceptual space that is shared between languages, suggesting it has a kind of universal ‘language of thought’.”
Anywho, there’s apparently a long way to go with this research. According to Anthropic, “it currently takes a few hours of human effort to understand the circuits we see, even on prompts with only tens of words.” And the research doesn’t explain how the structures inside LLMs are formed in the first place.
But it has shone a light on at least some parts of how these oddly mysterious AI beings—which we have created but don’t understand—actually work. And that has to be a good thing.
My favourite part of the day: commenting LLMentalist under AI articles.
this is one of the most interesting things about Llms that i have ever read
That bit about how it turns out they aren’t actually just predicting the next word is crazy and kinda blows the whole “It’s just a fancy text auto-complete” argument out of the water IMO
Predicting the next word vs predicting a word in the middle and then predicting backwards are not hugely different things. It’s still predicting parts of the passage based solely on other parts of the passage.
Compared to a human who forms an abstract thought and then translates that thought into words. Which words I use has little to do with which other words I’ve used except to make sure I’m following the rules of grammar.
Compared to a human who forms an abstract thought and then translates that thought into words. Which words I use has little to do with which other words I’ve used except to make sure I’m following the rules of grammar.
Interesting that…
Anthropic also found, among other things, that Claude “sometimes thinks in a conceptual space that is shared between languages, suggesting it has a kind of universal ‘language of thought’.”
Yeah I caught that too, I’d be curious to know more about what specifically they meant by that.
Being able to link all of the words that have a similar meaning, say, nearby, close, adjacent, proximal, side-by-side, etc and realize they all share something in common could be done in many ways. Some would require an abstract understanding of what spatial distance actually is, an understanding of physical reality. Others would not, one could simply make use of word adjacency, noticing that all of these words are frequently used alongside certain other words. This would not be abstract, it’d be more of a simple sum of clear correlations. You could call this mathematical framework a universal language if you wanted.
Ultimately, a person learns meaning and then applies language to it. When I’m a baby I see my mother, and know my mother is something that exists. Then I learn the word “mother” and apply it to her. The abstract comes first. Can an LLM do something similar despite having never seen anything that isn’t a word or number?
I don’t think that’s really a fair comparison, babies exist with images and sounds for over a year before they begin to learn language, so it would make sense that they begin to understand the world in non-linguistic terms and then apply language to that. LLMs only exist in relation to language so couldnt understand a concept separately to language, it would be like asking a person to conceptualise radio waves prior to having heard about them.
Exactly. It’s sort of like a massively scaled up example of the blind man and the elephant.
Yeah but I think this is still the same, just not a single language. It might think in some mix of languages (which you can actuaysee sometimes if you push certain LLMs to their limit and they start producing mixed language responses.)
But it still has limitations because of the structure in language. This is actually a thing that humans have as well, the limiting of abstract thought through internal monologue thinking
Probably, given that LLMs only exist in the domain of language, still interesting that they seem to have a “conceptual” systems that is commonly shared between languages.
It doesn’t, who the hell cares if someone allowed it to break “predict whole text” into "predict part by part, and then “with rhyme, we start at the end”. Sounds like a naive (not as in “simplistic”, but as “most straightforward”) way to code this, so given the task to write an automatic poetry producer, I would start with something similar. The whole thing still stands as fancy auto-complete
The other day I asked an llm to create a partial number chart to help my son learn what numbers are next to each other. If I instructed it to do this using very detailed instructions it failed miserably every time. And sometimes when I even told it to correct specific things about its answer it still basically ignored me. The only way I could get it to do what I wanted consistently was to break the instructions down into small steps and tell it to show me its pr.ogress.
I’d be very interested to learn it’s “thought process” in each of those scenarios.
How can i take an article that uses the word “anywho” seriously?
This is great stuff. If we can properly understand these “flows” of intelligence, we might be able to write optimized shortcuts for them, vastly improving performance.
