Mapping the Mind of a Large Language Model
Mapping the Mind of a Large Language Model
We have identified how millions of concepts are represented inside Claude Sonnet, one of our deployed large language models. This is the first ever detailed look inside a modern, production-grade large language model.
I often see a lot of people with outdated understanding of modern LLMs.
This is probably the best interpretability research to date, by the leading interpretability research team.
It's worth a read if you want a peek behind the curtain on modern models.
You're viewing a single thread.
This is a really good science communication article, it describes their work in clear terms (finding structures that relate to abstract concepts, seeing when they are activated and how strengthening and weaking them modifies outputs) and goes into the implications for it. I'm probably going to save this link as a rebuttal for the people who claim LLMs just predict the next word and have no concepts embedded in them.
I doubt that anyone saying that LLM are calculating next word solely based on previous sequence. It's still statistics, regardless of complexity.
Youd be surprised at the level of unthinking hatred around them, but even discarding that Ive seen it said often that LLMs have no internal model of what they are talking about as they are just next word generators. This quite clearly contradicts that interpretation.
concepts embedded in them
internal model
You used both phrases in this thread, but those are two very different things. It's a stretch to say this research supports the latter.
Yes, LLMs are still next-token generators. That is a descriptive statement about how they operate. They just have embedded knowledge that allows them to generate sometimes meaningful text.
Yes, but people forget that our brains, and therefore our minds, are also "simply" statistics, albeit very complex.
Yeah I found this kind of reductionist talk pushes people to overlook the emerging properties of the system, which is where the meat of the topic is. It's like looking at a living cell and saying "yeah well this is just chemistry".
Saying that it's "statistics" is, at best, unhelpful. It conveys no useful information. At worst, it's misleading. What goes on with neural nets has very little to do with what one learns in a stats course.
Most people don't know what Bayesian statistics are so you could say most people don't really get how machine learning works in general anyway. It's not misleading though as it perfectly sets expectations on what you're getting as output. It's much more healthy to general understanding of AI than anthropomorphizing very inflexible and limited models achieved thanks to technology that is seemingly in a plateau.
I would not expect almost human-like conversation on being told that is just statistics. I'd expect something like the old Markov chain jobs. What kind of knowledge leads you to have higher expectations?
Also, how does Bayesian statistics enter into this?
ELIZA from 1966 was enough to convince people that computer program they were talking to was human. People are now being sold on getting answers to their questions via natural language prompts and those answers are pretty much plausibly sounding sentences that happen to be right sometimes due to probability calculations.
Bayesian statistics is very different from what's being taught up until high school (at least here) and is foundational to earlier machine learning applications like spam filters. It's hard to imagine understanding what LLMs do without basics.
Those aren't the basics, though. That's how saying it's statistics is misleading. A Bayesian network is not a neural network.
Yeah, it's about as useful as saying that all of science is "just statistics". Which like, in a literal way, it's true. But science is still what forms the foundation of our entire civilization and base of knowledge.
Knowing that a blood pressure drug works is "just statistics", but you still take it if your blood pressure is high.
Yes, that's a valid comparison. It's worse with neural nets, though. Much of machine learning is literally applied statistics. That is, a program is written that applies statistical methods to data and then adjusts its behavior. So, saying that it's statistics has the potential to really send people down the wrong track. Many of the "human hallucinations" about AIs result from confusion about this.
Yes, good topic, good research...
( you have a few typos : intobthe ... into the, predicr ... predict, im ... i am. )