Take sick days? Are they stuck in 2019?
Excuse me, haven’t we already seen tech companies going back to using junior programmers and boilerplate code because the LLM tokens are more expensive than just hiring people?
That’s why taxpayers are paying for massive data centres that poison their neighbourhoods.
More data centres means token become cheaper and more jobs get replaced.
Tax payers literally paying to reduce the number of jobs for their children to work in the future.
haven’t we already seen tech companies going back to using juniors
Not really no. There might’ve been few outliers.
Companies are trying to spend their money more efficiently right now (as in, “make the genie work peasants, but for pennies”).
Source: ehh, I still work for one.
So we’d go back to indentured servitude?
Milord deserves all his riches! Milord works so hard and smart unlike me, please milord, bless me with your presence!
And how will these promising new “HOTL” have money to pay for entry level jobs without having entry level jobs?
And why use “HOTL” meaning “human in the loop”? What’s wrong with HITL?
Probably to differentiate it from hardware in the loop, which is what HITL typically refers to. That or she’s just not very good at this. Probably the second one.
I definitely read that as “humans out of the loop”
There is something rotten about that writing. I can’t read the drivel more than a paragraph at a time. It is like every sentence should have it’s own paragraph and the missing sentences have left the flow of thought all fucked up. I have spent plenty of time around cokeheads and such and their train of thought takes the scenic route compared to this crazy train.
I think Jenny K should do this.
Her essay is almost unreadable. Please tell me how you think a recent college grad, probably with school loans, is supposed to pay for their own “on the job training”? I’d love to see your magical economics at work.
Through behind closed doors activities that can’t be reported officially, I imagine. Warm bodies can be used for all sorts of underhanded things.
Top-tier shitposting
- 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.
She’s wrong about the advantages of AI, but she’s right about how companies are going to see it. It’s true that they’d rather deal with AI than an intern, and it’s true then that internships will fall onto the universities to coordinate like rotations as part of an expensive education system.
It’s also true that kids are learning how to use AI on their own, and are losing their ability to learn from any datapoints, because why do that if the magic box will give you an answer that sounds correct enough? Ultimately, it doesn’t matter how useless AI is if everyone acts as if it’s as useful as this lady thinks it is.
This was already a thing when I went to university over a decade ago. Partnerships with local businesses where university students basically do work placements, completing low-stakes corporate projects as assignments.
I know this goes against expectation but it was actually really good. I think the difference was that the business actually cared about us, taught us things, and even extended a job offer to people who did well.
I dunno if that program is still good today, probably not.
Yeah, local students do projects for local manufacturing plants to find efficiency improvements. I have second hand knowledge (girlfriend)
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)
50 First Prompts
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.
I would also like to add another key difference between human learning and machine learning. When I, as a person, read a book, I can conceptually understand the story. A machine would digest that same book by, for example, counting the numbers of times each word comes after another and assigning probabilities to occurrences or words or basic sentences. The major differences are still being explored. We don’t fully understand how humans learn, fundamentally. We know somewhat mechanisticly that neurons share and process chemical signals, but we don’t know how that information is interpreted or stored. We don’t know the equivalent of a computer language of the neurons (or other tissue types that may also learn). We know that we require input to learn, as do computers, the similarities kind of end there. That said, we definitely learn differently from how LLMs “learn.”
That’s not true at all. We know how learning and memory work?
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.
As you converse with it and it responds, part of the input is just your entire conversation up to that point, and its responses, plus your new prompt. That’s it that’s how the memory works.
And for this you pay an increasing amount for tokens.
Michael Reeves did an interesting piece on it: https://youtube.com/shorts/WP5_XJY_P0Q
Yeah I understand that much, my question was how does that personal record differ from what it already ‘knows’.
In my experience, it seems to be limited to previous conversations, and it may be limited in length.
I pretty regularly delete old conversations, partly with the guide of privacy which I am sure does not actually delete or exist, but also for my own sanity of looking back on projects. And I don’t even use AI that much, it’s all free tier, mostly Claude.
That said, it will often get “lobotomized” and “forget” things after I delete the source conversation.
It does remember basic things. Like I use it a lot to help set up Docker containers, so it knows I have a Linux machine and a webserver and a Windows Machine. It knows the Linux box runs Docker and is named Squall. It knows these sort of, generic things, even across conversations. I can open a new chat and say “I want to set up a Wordpress sandbox” and it will ask like “Would you like to use Docker on Squall or run it outside a container with Apache.”
Well, no. Models are not “self-learning”. This is unrelated to size of the corpus. In fact, it was proven times and times again, that poisoning a model is not a costly proces.
Most harnesses that people use to interact with llms have memory and skills.
You can “teach” them a skill which is sort of like a macro so you can say “deploy to dev” and it’ll look up that skill and execute it without any extra input needed.
For memory it’ll keep files on your local of what it’s seen before and can pull from the information in those files if necessary. So if you always ask it to label reports in a certain directory “Midwest expansion project” the harness will add that to its memory files and start to label it automatically.
Good luck finding the future mid-level class of your company when all the lower rungs are filled with AI. Who you going to promote? AI Albert?
“Cooper AI Co-founder & CEO”
Why would you expect her take to be anything but brain-dead?
Exactly.
Im pretty sure graduates aren’t beating down her door for the opportunity to work for her.
Sniffing her own farts too much.
It’s 2026 and yes, I can imagine that there are some rich college kids who think paying for an internship with her would be a smart move.
They’d get a chance to build up contacts and see how she operates.
Telling the Queen how good her farts smell has long been a path to success.
But only for those incapable of learning to live off lentils
Idk there’s a bunch of ‘contacts’ I know, and I’d think someone is a shitheel if another shitheel was talking them up
Ummm, it’s 2026. What exactly makes you think that being scum is somehow a barrier to success?
It’s not, I’m not the one checking references.
I’m pretty underwhelmed before even clicking…
Because, as we all know, the only people who can be adequately replaced by the inference-free prediction machine are business idiots.
Everyone else has a real job that actually requires real intelligence instead of prediction algorithms.
Funny enough, though their jobs would be the easiest to replace, they are “important” in that they can be held legally responsible for business decisions (Responsible Corporate Officer doctrine). That’s partly why they get tens of millions of dollars - so they can take the fall when shit hits the fan (either they fuck the shareholders or majorly break the law).
This is known as the “Scapegoating Premium” - you are basically paid for the risk of taking on a very public, high liability/visibility position. The corporation pays their legal fees and has insurance to help with this, but it’s compensation for the possibility of legal and reputational damage.
Very little risk. Laughable in most countries. C-Suite accountability translating to truly damaging punishment, like jail time, is almost nil.
Yeah, I’ll believe corporations (or c-suites) are people when Texas executes one.
Oh for sure. But regardless, the potential for reputational harm and liability are part of why named execs get these ridiculous sums.
No evidence, just mindless opinions based on nothing.
Here’s my take: In the future, Tomatoes will become sentient from genetic engineering and will wish to overthrow their masters. Only gen z, alpha and AI will be able to stop this…Attack of the Killer Tomatoes. I’m smart and realistic.
What a fucking bunch of drivel. Without students and fresh grads our company would have been bankrupt a long time ago. They are the ones who bring a fresh perspective to the company (we engineer machines for the beverage industry) because they haven’t been pushed into a specialist position yet, that breaks any kind of creativity. They are the people that spark a lot of our innovations. If we lose that, the company won’t survive for long. AI can’t do that because AI isn’t fucking creative. Fuck AI and fuck this dumb piece of shit for their dumb fucking take on this.
Sorry for the incoherent swearing, I’m drunk.
Love you all, have a great day!
She’ll be looking for sympathy after the a.i. bubble bursts and she’s bankrupt.
There is NO legitimate reason to endure the fresh graduate.
So many incredibly entitled people have never felt the obligation to give to others that which was given to them. These people should be shunned by civil society.
