- is an insanely fast learner
This person very clearly does not understand, even at a basic level, how llms work. They are completely incapable of “learning” outside of training, and training cannot happen while you are using them - in no small part because they require MILLIONS of tokens of data (minimum) to be able to recognize patterns.
Meanwhile that theoretical stupid kid she thinks should be paying her masters to have her teach them can learn instantaneously from even a single datapoint.
I’m told that agents these days keep a scratchpad of notes to themselves, so an agent can (kind of) learn about things by making a note about it which it will still have access to later, even outside the formal context window.
It’s going to be far more limited than the kind of training you’re talking about, but it is something which an individual agent can do
That is sort of true, but it still isn’t learning. Basically, it can store information usable by other sessions. But to do so, it just processes that information the same way as a prompt. So less like learning and more like just tacking something on to its context window. So, it never “learns” that information, it just tells itself that thing every time it processes any prompt.
(don’t downvote MartianSands’ comment - it’s a good point. Shame on you)
Explain the distinction in more detail, and why it matters.
Say you had a version of chat GPT that hadn’t been trained on a corpus that included enough information about the number of Rs the word “strawberry” contains, didn’t have workarounds for character counting, and has context sharing.
You ask it “How many Rs does Strawberry contain?”
It burns a bunch of tokens, returns “Strawberry has 7 Rs. No wait, that’s not right, strawberry has 5 Rs. No wait…” etc.
You say “Strawberry has 3 Rs. How many Rs does Strawberry have?”
It says “Strawberry has 3 Rs.”
You ask “How many Rs does this exact string contain: ‘Strawrberrrry’?”
A new session using shared context responds “Strawberry has 3 Rs.”
You say “No, it has a different number of Rs. I didn’t ask about Strawberry this time”
It responds “Sorry about that! You didn’t ask about Strawberry, you asked about Strawrberrrry. Strawrberrrry has 7 Rs. No wait, that’s not right. Strawrberrrry has 5 Rs…”
LLMs, at their core, use models that have computed lexical & semantic similarity to predict words (really they perform contextualized vector transformations, but let’snot get into that). When they “learn”, they just do this more effectively using more text or using the same text more efficiently. Next word prediction is not the same thing as learning concepts, like the concept of numbers, non-numerical things being ascribed numerical values (ironically, since LLMs function by turning words into numerical vectors), or counting.
When it responds “Strawberry has 3 Rs” it isn’t because it knows what that means or how to gain that knowledge about other words. It is merely parroting back what you’ve told it because “Strawberry has 3 Rs” in its shared context has very close lexical similarity to your query “How many Rs does Strawberry have?”. It also parses Strawberry as [“straw”, “berry”] encoded into numerical values representing their relationship in the corpus’ vector space (e.g. {[ .420, -.67, .67], [.420, .69, -.69]}) - so unless it has instructions to further break those tokens into characters, it does not have the capability to count. You can actually do some weird math using these vectors and their relationships (their distance apart in the 3D vector space, angles between vectors, etc actually relate to the semantic content of the tokens), but you lose granularity such as the number of letters in a token when you look at words this way.
Learning is more than just computing lexical similarity. Modern LLMs mimic reasoning by expanding the query with intermediary tokens, basically creating a temporary scratch pad of related words on the same vector, but don’t actually reason (this is known as chain of thought). Learning is partly computing lexical similarity, but it is also about extrapolating concepts from facts and inferences, applying those concepts to novel simuli, etc. LLMs really don’t do reasoning, which is part of learning.
Counterpoint:
“Teach” the LLM the proper procedure for counting the Rs and tell it to commit that to a file (or as a tool call). Then from there it will suceed 100% of the time.
It doesn’t matter if it’s true learning or not. When you teach an employee to do something you give them the procedure and have them do it. If AI can be fed a procedure too, it will do it just fine even if it’s not in its true training. Does it matter if it’s consuming the procedure as prompt context everytime? Hell a lot of humans do read the procedure every time when it’s complex enough.
What the example shows is that you cannot “teach” the llm how to count the letter R, because LLMs don’t work that way.
The AI company would have to solve the problem another way, let’s say by recognizing that the user is asking for letter-counting, and pass that off to a different kind of algorithm that can count letters.
Yes, and that’s called a harness, which everyone uses these days. The harness increases perceived intelligence (or accuracy) by absurd amounts. You can “teach” (or the equivalent of) LLMs anything with a custom harness.
Forgetting the handwave of LLMs and AI - this is a classic problem for a long time. Apprenticeships have existed for a long long time, and the problem of needing skilled labour in the future, but only having unskilled students now is an old one.
In a sense, she’s right. Companies need to be compensated somehow to take on apprentices - otherwise you just intern at a company, take way more resources in training than benefit you provide, and then you might fuck off to a competitor all trained up.
But this doesn’t have to be “pay for internship” (which is essentially what a trade school is anyway - “you mean I pay money to do welding?”, “Yes because you do it so badly at first that you have to do it on practice metal and it’s totally unusable and we have teachers helping you practice until they say you’re good enough”). This could be making agreements to stay on after your training period, or join a body of some sort that you pay dues towards that sort of acts as insurance against this sort of thing, and the body handles the risk of your training (i.e. a plumbers union or some shit).
So that’s not how this community works I found.
When they are fed obvious ragebait they want you to rage with them. Multiple paragraph arguments need to happen in a separate post which will have barely any discussion or votes.
This is just rage bait.
Well you did come to the bring your pitchfork community (:
The funny part is AI and LLM don’t even have to do anything actually intelligent or malicious and we’re already up in arms about a lunatic on linkedin of all places.