Hacker News Reader

Comment by CGMthrowaway | Hacker News Reader

>INSUFFICIENT DATA FOR MEANINGFUL ANSWER

Boy, it sure would be nice if real LLMs were capable of giving an answer like that.

Living in South America a bit really showed me this. I think it's a cultural thing here but someone will always give you an answer, even if it's wrong, confidently. It was hard for me at first- I am usually the first person to say "I don't know" (often followed by "but let's slow down and find a good solution").

This was similar to my experience running a software team in India (I'm an American) a couple decades ago. I had to learn not to ask yes/no questions because the answer would always be yes.

It's a long-standing jodke that AI stands for "Actual Indians".

From real life: - Is it done? - Yes! But not yet!

On a long enough time horizon, with sufficient multiverses, all things are done!

Talking about South America as a homogeneous unit is… weird. Even neighbouring countries speaking the same language can be entirely different in this regard.

I agree (and I don't normally generalize like this, so I apologize). I've spent most of my time in Peru but noticed it in neighboring countries as well.

can speak from personal experience that it's the same culturally in colombia

That also is my perception, from Brazil. There is even the concept of "cordial man", coined by sociologist Sergio Buarque d Holanda, that is connected to this

https://en.wikipedia.org/wiki/Cordial_man

i remember >20 years ago going to the bus station somewhere between RJ and SP, and asking the best way to get to Iguaçu

  - it's difficult
  - ok fine but how
  - it's difficult
  - right i'll see that but how
  - it's difficult

then it dawned on me this meant get away you fool :D

"You can't get there from here"

This reminds me of a great adventure. A long time ago I was travelling through Brazil, in the Amazonas state. I was in Porto Velho and needed to get to Manaus quickly to catch a flight. The boat that would take us there (as I had found in the Lonely Planet) was present on the quay. But the captain didn't feel like going. If I remember correctly, he was waiting for more people. We needed an alternative route quickly.

The captain told us that if we took the bus to Humaitá (a smaller provincial city), we could take a smaller boat that would take us to Manaus. But he warned us that the boat only goes once in three days, and that it would leave soon. The last bus to Humaitá would also leave from Porto Velho promptly.

Despite this flimsy instruction, we didn't see any alternative. So we went. With great luck, we caught the bus and made it to Humaitá (I still have a picture of the boat river transfer the bus took: https://bashify.io/i/CdNcLf).

Our time in Humaitá was surreal. When asked about a boat to Manaus. Everyone told us a different story. There was no harbour (hydroviaria) personal. One person told us the boat to Manaus was named "Caçote". Another person said the boat was named something else. Then someone said it had stopped ferrying years ago. No, we heard from someone else, it would come in 5 hours! Yet another one said it was tomorrow. Someone else felt sorry for us because it had just left. I felt like I was in a (difficult) point and click adventure. There weren't a lot of people in town close to the river, so we ran into the same people from time to time. They would often give different answers to the previous time.

No one was willing to tell us they didn't know. Not a single person out of the 20+ we must have talked to.

In the end, a boat arrived. It went in the direction of Manaus. The captain said that he would only go up to Auxiliadora, and that was still a long way from Manaus. Once again without alternatives (going back to Porto Velho would surely mean we'd miss the flight), we chanced it. In the hopes that getting closer was worth it.

When we arrived at Auxiliadora, it was the smallest inhabited place I had ever been to. Perhaps about 13 houses. Some fishermen. No passenger boat would come for days, they told us. Not to take us to Manaus, neither to take us back. The fishermen had boats, I tried to offer them money so they would take us further. But their day was at an end and they wanted to relax, regardless of what I offered them (we were on a tight budget, but I was desperate enough to offer a significant chunk of a monthly wage, no dice).

Then we found out that on the boat with us, a woman had come who was in a similar predicament as us. She was Brazilian, living in MT and wanting to visit family in Manicoré (which was bigger, and closer to Manaus). Exasperated, she ended up convincing her family to come and pick her up with a speedboat. We hitched a ride. We were very thankful.

When we arrived in Manicoré, I felt like exploring the place. It looked so different from anywhere else I had been, like something out of a movie. But I couldn't. The docks were little more than a collection of wooden jettys (trapiche) that ran everywhere in criss-cross fashion. In order to even get to the quayside we would have to pass through many other boats. In the first one we went through, the captain walked past and I asked him whether he knew of a boat going to Manaus. He signalled where to put my bags. We were leaving.

We reached Manaus in the nick of time.

I love this story, and that time. These anecdotes definitely triggered my memory.

> someone will always give you an answer, even if it's wrong, confidently

its common playbook for corporate self-development in NA.

I've experienced similar with some Southeast Asian cultures as well.

I'm a patient person, but it can be frustrating to have to endure 10 minutes of verbal diarrhea that eventually results in a "no" or "I don't know".

Over the last couple of years, I've spent a lot of time in Indonesia. By the time I got used to their way of communicating, I questioned my own reality, perception and sanity. I even put a thought it's some very passive way of gaslighting foreigners. It seems it's just how they like to do it here.

I'm genuinely curious if this is a thing with roots in Spanish culture? Because there is strong Spanish influence in Philippines and South America.

I don't know any Spaniards but I do know Filipinos and the confidence projection is a real thing. The Filipino IT guy confidently declared that my OnePlus Android phone wasn't certified for the software he was trying to install and was getting errors. It is a bog standard application that can be installed on any modern Android phone but the level of confidence he projected, just because he didn't know OnePlus as a brand, made me doubt myself until I turned on the critical hat and pushed back a little with alternative approaches, which solved the problem.

Is South America populated by LLMs?

But I kid, I have a friend who's the same way. He's an Austrian who grew up in Chicago and was in the army.

I have considered the phenomenon. I somewhat disapprove but I can also see the advantage of always presenting a confident face

They can do it, it's just not "by default", they need to be prompted to do it. So at least the danger is manageable if you know what you're doing and how to prompt around it.

Not really. They're still non deterministic language predictors. Believing that a prompt is an effective way to actually control these machines' actual behavior is really far fetched.

They com like that from factory. Hardcoded to never say no.

The thing is that they are completely incapable of meta-cognition. Reasoning models don’t show their actual reasoning at all.

Right — they're not reasoning, they're generating text that statistically models reasoning. Anyone who says differently is selling something.

That is what a base model does. After RL it is a very different thing, and anyone who says they know what it is, is naive or dishonest. These things are grown, not made, and we really do not understand how they work in many important ways.

Yeah, but they’re not magic; we can still do experiments and see what happens. Anthropic did a lot of work on this and showed that they’re not accurately describing their reasoning process.

Of course, the fact that they have to do that proves my point.

Humans are also not magic; we also do experiments on humans, and humans also do not accurately describe our reasoning process.

Lot of confabulation going on in the moist blob of electrochemistry found encased within the hydroxyapatite crystal cage we call a skull. Why is it that things which rhyme, or get repeated a lot, seem more true? How come some of us suffer uncontrollable seizures from flashing lights?

We have to study reason to get any good at it, we absolutely suck at this without training. Our natural state is illiterate, innumerate, and illogical.

The part of our (all animals, not just humans) intelligence that is most magical*, is how efficient we are with few examples, not reasoning.

* Questions about consciousness will have to wait until we can agree which of the 40+ definitions of the word actually answers the question we care about, and then also we figure out how to actually test for whatever that is.

As the meme goes, "they are the same picture".

Language has reasoning encoded within it.

It certainly does. But so too do complex neural network functions, as do attention mechanisms.

They're not hardcoded to never say no, but some of the models were trained to be "yes men" because their creators thought it would be a good property to have. GPT-4o for example.

Not believing that a prompt is an effective way to actually control their behavior is obviously incorrect to anyone who's actually used these things.

It's not a guaranteed way to control their behavior, but you can more than move the needle.

yeah that distinction is pretty important, and in general that guy I believe IS making the point - if you can not control it with guaranteed outcomes - you cannot control it.

You can't control it any more than you can control a draw from a deck of cards, but you can absolutely control the deck of cards that you choose to draw from.

The problem is that nobody really does that? Like, as far as I'm aware, even simple stuff such as not considering tokens that would result in a syntax error when writing code isn't being done.

magicians can probably make you change your mind on the former

That's silly. My car is not absolutely guaranteed to turn left when I turn the steering wheel left, but you wouldn't say I can't control my car on that basis.

Steering an LLM with a prompt is way less reliable than steering a car with a steering wheel, but there's still control. It's just not absolute.

if your car doesn' turn left when you turn the steering wheel left, the problem is that the car is broken, if an LLM does something unexpected after you gave it instructions, that's possible when the LLM is functioning entirely correctly.

Nothing in this world is guaranteed. That doesn't mean it's uniformly random either. LLMs can still do something unexpected if you give them clear instructions, but that doesn't mean it'll be arbitrary and unpredictable in scope. The same way C/C++ undefined behavior technically means program can give you nasal demons, but in reality it won't do anything unusual (like format your C:/ drive) unless someone purposefully coded it to do that.

This is all going to flash through your mind when your car mysteriously doesn't turn left. I would prefer to think of machines as things with defined outputs and failure is failure, more than as fluffy little kittens who might do the wrong thing, if the consequences are going to fall on someone who doesn't deserve it.

The word most relevant to this conversation is “influence.” Influence is possible and users observe it and use it to increase margins of useful outcomes. “Control” is incorrect.

> non deterministic language predictors.

Non?? Only those with sh*tty code, surely.

There's nothing inherently non-deterministic about inference.

"Just don't accidentally forget to do the thing that makes it safe" is not a very effective strategy for something that so many vested interests are trying to push into all corners of society. If it's so easy to misuse it, then it shouldn't be used in any context outside of where there are no major consequences for bad output and there's amble opportunity and ability to validate it

There are a lot of humans who refuse to give that answer, too

I don’t have to spend dozens if not hundreds of dollars a month to talk to most people in my life lol

Do you have to talk to LLMs?

Another way to say the same thing: "to talk to most people in my life lol I don’t have to spend dozens if not hundreds of dollars a month"

There are lots of free LLMs, but you (often) pay for the good ones.

There's lots of people with lots of opinions, but you (often) have to pay for the good ones.

According to HN, every employer, and general social chatter, apparently yes.

Well, speaking from what I hear and see, employers want you to start using it so that you can be more productive. They've been sold this tool and want you to learn it so that your output will grow.

That's not an unfair take, I think. Again, just IME, they expect too much because the tool is oversold: it does not deliver that well. And we always hear, this new model is so much better, it's tiring.

I think we should all learn to use LLMs but we should still carefully review what they did. And that is what the employers don't quite get: the review still takes a lot of time. So, gains are not 10x but more like... 10%? Maybe 50 for boiler plate. Still gains are there, I guess.

> they expect too much because the tool is oversold: it does not deliver that well.

And unfortunately a lot of people will say it’s their reports’ fault for not properly utilizing it (even as they barely use it) because otherwise they would have to admit that they bought a tool without any plan for how to deploy it. So regardless of what is or isn’t a fair take, the results are the same. We are burdened with utilizing a thing whether it is useful or not and the results are generally not what is measured, but rather “are you using it?”

I’m just glad I work at a company that has more reasonable expectations and has been very slowly, thoughtfully rolling it out to individuals at the company and assessing what is and isn’t good for. They are interested in getting me in line, but as somebody in video production to be perfectly honest the use case for Claude is a bit tricky to navigate. We don’t write a lot of scripts and I already have bespoke software for organizing/maintaining footage that isn’t on a subscription basis. The work I’m also doing doesn’t call for these speed-editing solutions that generate tik tok chaff. All our stuff is hours long and it’s high volume. Any video-centric AI service costs an arm and a leg.

I do think it could be useful for writing some terminal scripts and such, but as far as a daily tool we are still scratching our heads and thinking about it. But it’s nice to be able to do that without somebody saying “why aren’t you using it?” every meeting.

Why are employers so incompetent to just believe and cargo cult any business trend to come along? Shouldn't they do research first before making wide, sweeping changes in work policy?

I know I'm shouting into the void, but seriously.

Book an appointment with a Psychiatrist, it’ll cost more than a months cc subscription for sure

This “joke” was such low hanging fruit I went and added “most” after I initially wrote the comment in anticipation of it.

This continues to be the most tiring response to any criticism of LLM output. It's pretty much guaranteed to show up at this point. I guess with similar enough input tokens, we're guaranteed the same output...

You’re absolutely right! I do have insufficient data for a meaningful answer. This is not an *insightful prediction* — it’s *Dunning-Kruger masquerading as qualified intelligence*

No Information before. No information after. This is not a failure — it's narcissism as a service.

YaaS — yes-men as a service

Did a human write this?

I would guess a real human, one with a good sense of humor at that.

Woosh

Exactly!!

I've been trying to work on a new LLM code editor that does just that. When you instruct it to do something, it will evaluate your request, try to analyze the action part of it, the object, subject, etc, and map them to existing symbols in your codebase or, to expected to be created symbols. If all maps, it proceeds. If the map is incomplete, it errors out stating that your statement contained unresolvable ambiguity

I think there is a real benefit here, and it might be the actual next beneficial grounded AI sustainable use in programming. Since I the current "Claude code and friends" are but a state of drunkenness we fell into after the advent of this new technology, but it will prove, with time, that this is not a sustainable approach

At the time of this writing, the prevailing thinking with "artificial intelligence" was that we'd encode every Fact we know and every rule of Logic, and from there, the computer would make new discoveries. Todays AI researchers would call this "symbolic" AI, compared to the "neural" AI powering LLMs. They're like two different worlds.

LLMs are just generating text, they don't know anything. They can't assess whether there is enough data for an answer. When you add a follow up prompt "This is wrong, why did you lie?" only then is it able to generate text, "I was wrong, I'm sorry," and so forth.

Did Asimov’s idea of AI revolve around data retrieval? I’ve read that even human intelligence isn’t necessarily remembering things, but being able to traverse our knowledge and find that idea or thought quickly.

They can read context and with fairly high accuracy say whether that context contains enough information to answer a posed question. They cannot (and we cannot for them) introspect their own weights to say whether their weights already encode information sufficient to answer a posed question.

This is exactly like a lot of customer service, or technical support.

It seems that they are loath to tell anyone “no”, or that something can’t be done, or that an app doesn’t have a feature or can’t be used in a certain way. Especially when a feature has been removed for security reasons.

In fact, it gets so crazy that I simply cannot get a straight answer out of somebody and if I persist in my line of questioning and they become evasive or vague or I just can’t get a straight answer for long enough, ultimately, I suspect that the answer is “no”, and that they're simply not allowed to tell me, and they're paid and trained specifically to avoid uttering the “n-word”.

In my first job, as a network operator, my supervisor admonished me, and said “we must never tell a customer that we don't know something”. He said that we should tell the customer that “I will go ahead and find out for you, and get back to you on that”.

And that is kind of the kind of slippery non-answer I often received in my most recent job, that some manager or supervisor would “look into something” for me and “get back to me”. But the ‘getting back to me’ part never happened, and I began to suspect that it was a platitude meant to satisfy me enough that I would shut up for a while, and stop pressing the issue.

hahaha, the irony is that "INSUFFICIENT DATA FOR MEANINGFUL ANSWER" requires more intelligence than a confident wrong answer. you have to know what you don't know. current LLMs are optimized to always produce output, which means they've essentially been trained out of epistemic humility.

Asimov's Multivac at least had the dignity to wait.

Yes but it waited until the end of the universe. Its annual KPI and OKR performance was abysmal.

I reckon that’s how we know we’ve hit ASI.

As measured by #_no_answer/(#_incorrect + #_no_answer) the top current models can do it 60-70% of the time (Grok 4.20 is the best with 83%): https://artificialanalysis.ai/evaluations/omniscience

2061, mark the date

Just add a skill to Claude

I just came from reddit and seeing this comment, looked for "controversial" sort option instinctively.

Maybe hackernews is becoming reddit...

Reply on news.ycombinator.com