A thread for unboxing AI
The rapid progression of AI chatbots made me think that we need a thread devoted to a discussion of the impact that AI i
I'm thinking you're probably right in theory. Given enough monkeys and keyboards along with the computational power to check for coherence, eventually the Pythagorean theorem, calculus, etc. would flag as a hit. Same with something like mouthwash if it hadn't been invented by combing words in the dictionary and seaching for correlates and need, like posts about bad breath. Now
This is where the idea that LLMs are just guessing words can be very misleading.
Comparing it's search is like comparing survival of the fittest to random guessing the answer.
I know LLMs hallucinate regularly, and I've gathered they are not really candidates to become AGI, or rather, they aren't real "artificial intelligence".
Dunno. I'm skeptical that LLMs will be sufficient but I also don't think the hallucination objection is a serious one to the LLMs in general
My guess is that we will se the real action start soon with robots wandering around modeling the world and making decisions. They will incorporate LLMs or maybe will just be considered to be extensions of LLMs. They're likely to have their autonomy as well as connecting to distributed and/or centralized systems.
I know LLMs hallucinate regularly, and I've gathered they are not really candidates to become AGI, or rather, they aren't real "artificial intelligence".
Hallucinating is a sign of thinking. Basically, humans are hallucinating all the time 24/7.
This is where the idea that LLMs are just guessing words can be very misleading.
Comparing it's search is like comparing survival of the fittest to random guessing the answer.
Sure, not guessing, but they’re still operating through material implication, not genuine inference. That won’t change no matter how much information they take in or how fast or accurately they correlate it.
For example, if all an LLM knows is that “If pigs can’t fly, then Trump is president” is true, and it’s then told that Trump isn’t president, it will conclude that pigs can fly. And that sort of issue never goes away, no matter how much data it acquires. It's baked in.
I doubt it. It's not even a correct inference but even if it was that doesn't mean it would reach that conclusion.
Dunno. I'm skeptical that LLMs will be sufficient but I also don't think the hallucination objection is a serious one to the LLMs in generalMy guess is that we will se the real action start soon with robots wandering around modeling the world and making decisions. They will incorporate LLMs or maybe will just be considered to be extensions of LLMs. They're likely to have their
Tell me you don't have any technical expertise without telling me you don't have any technical expertise.
Jeeeeeez. Just stop it.
Sure, not guessing, but they’re still operating through material implication, not genuine inference. That won’t change no matter how much information they take in or how fast or accurately they correlate it.For example, if all an LLM knows is that “If pigs can’t fly, then Trump is president” is true, and it’s then told that Trump isn’t president, it will conclude that pigs can
As Bob ask so nicely. simplifying to being told X
Excellent and subtle question — this goes straight to the boundary between **logic** and **epistemology (knowledge)**.
Let’s unpack it carefully.
---
1. In **formal logic**
From the statement
> “I’m told x”
> you **cannot validly infer**
> “x”.
That’s because “I’m told x” only tells you that **someone asserted x**, not that **x is true**.
Formally:
* Premise: Told(x)
* Desired conclusion: x
There’s no logical rule that allows that step — it’s a *non sequitur* (it doesn’t follow).
---
2. In **everyday reasoning**
We *often* treat “I’m told x” as weak evidence **for x**, if we trust the source.
That’s an **inductive inference**, not a **deductive** one.
It depends on background knowledge like:
* How reliable is the source?
* Do they usually tell the truth?
* Is there corroboration?
So, “I’m told x” → “x” is **not logically valid**, but it can be **rationally plausible** depending on context.
---
3. Summary
| Type of reasoning | Can you infer “x” from “I’m told x”? | Notes |
| ----------------------- | ------------------------------------ | ------------------------------------------- |
| **Deductive logic** | ❌ No | There’s no rule allowing that inference. |
| **Inductive reasoning** | ⚠️ Sometimes | Only if the source is reliable or verified. |
---
So, in strict logic: **no**.
In human reasoning: **maybe**, but it depends who’s telling you.
LMAO
Chezlaw: "Oh noes, someone accused me of having no clue what I'm talking about, better post a confused, horribly formatted AI slop snippet, that'll show em!!!!"
Oh don't be sorry, I'm sure it all makes sense in your head.
It's my fault, I am way too impatient with clueless, uninformed people...
Yes I can understand the difference between 'told X' and 'X' in logic and inference. As so clearly shown, so can LLMs
You probably could too
I agree with that. But that's not quite what I meant. What I meant is what happens further down the material logic stream ie after it’s already correlated two true statements.
x → y
¬y
∴ ¬x
2024: x = pigs can’t fly; y = Biden is president.
2025: y = Biden is not president. Therefore, ¬x = pigs can fly.
With material implication, it’s simply the correlation of two facts that coexist in time. There’s no genuine inference between them, unlike:
If (x) Biden won the election, then (y) Biden is president.
(not-y) Biden is not president.
Therefore, (not-x) Biden didn’t win the election.
So I'm not disputing the logic upstream. We know the absurdity of "pigs can fly":but to AI that phrase is just x or not-x.
That’s the difference between correlation and inference, not the logic. LLMs operate on the former: they detect and extend patterns rather than truly understand causal relationships. Yet that same limitation is also their strength:
[QUOTE=AI]For decades, biologists struggled to predict a protein’s 3D structure from its amino acid sequence. Then AlphaFold came along — not by reasoning like a biochemist, but by correlating millions of known sequences and structures. It didn’t “infer” chemical laws; it recognized statistical regularities so subtle and high-dimensional that no human could consciously grasp them. Yet the result was the same: a breakthrough understanding of biological form and function.[/QUOTE]
But my point is that we know this issue is baked into LLMs, so there will always be a non-zero chance they’ll direct us to a pig farm when we’re looking for a flight. So in mission-critical situations, we’ll always keep human eyes on it to verify that its syntactic or symbolic reasoning lines up with our semantic expectations.
Tell me you don't have any technical expertise without telling me you don't have any technical expertise.
Jeeeeeez. Just stop it.
LMAO
Chezlaw: "Oh noes, someone accused me of having no clue what I'm talking about, better post a confused, horribly formatted AI slop snippet, that'll show em!!!!"
Oh don't be sorry, I'm sure it all makes sense in your head.
It's my fault, I am way too impatient with clueless, uninformed people...
That's cute but won't help you with your problem that is: not even understanding the basics of an LLM
Look at you, all clueless, acting like you know something. So sad.
I'm not seeing any evidence here that you know more about this stuff than Chezlaw does.
That's your read. Mine was that he took my “material implication” point literally and thought I was talking about pure formal logic. That's why I agreed with him on what AI said and then went on to explain what I actually meant.
But that's my read. Just like my read on your responses is he tied you into a pretzel at some point and you haven't gotten over it.
That's your read. Mine was that he took my “material implication” point literally and thought I was talking about pure formal logic. That's why I agreed with him on what AI said and then went on to explain what I actually meant.
If you think he knows what he's talking about, I have very bad news for you.
But don't worry, there are literally millions of you, thinking you got this whole AI thang figured out so well that you can make informed predictions lmao..
If you think he knows what he's talking about, I have very bad news for you.
But don't worry, there are literally millions of you, thinking you got this whole AI thang figured out so well that you can make informed predictions lmao..
What he's talking about with survival of the fittest is what we see with A/B testing for optimization.
That has prediction value.
I agree with that. But that's not quite what I meant. What I meant is what happens further down the material logic stream ie after it’s already correlated two true statements.x → y¬y∴ ¬x2024: x = pigs can’t fly; y = Biden is president.2025: y = Biden is not president. Therefore, ¬x = pigs can fly.With material implication, it’s simply the correlati
There's so much to unpack in all this. Is part of your argument that LLMs are restricted by their training data because they don't incorporate new information very well i.e doesn't infer new answers that are beyond or contradict the answers in the training set?
re your time point I would disagree that LLMs cannot do temporal logic. We tend to assume a lot of temporal modifiers i.e biden didn't win the election means he didn't win the last election when of course he did win the election (and was president) if we refer to a previous election. I' don't think this is a fundamental problem. Bob might have disgreed one day soon.
Did you eat a lot of paint chips when you were a kid?
There's so much to unpack in all this. Is part of your argument that LLMs are restricted by their training data because they don't incorporate new information very well i.e doesn't infer new answers that are beyond or contradict the answers in the training set? re your time point I would disagree that LLMs cannot do temporal logic. We tend to assume a lot of temporal modifiers
“l * w = a”
We know exactly what that means, but to an LLM it’s just a string of 5 meaningless symbols. Even if we define what “w” means to us, “width” is still just another string of 5 meaningless symbols to it.
LLMs are just correlating symbols; symbols that only mean something to us in the real world. It’s symbolic logic that lacks any semantic grounding at all.
That’s why they can’t genuinely infer: without any semantic grounding, they can’t understand what the symbols refer to in the real world. And that’s the inconceivable challenge of AGI: how to create, or even mimic, having semantic skin in the game of real life.
But as we were talking about. maybe human inference itself is just a heuristic we’ve evolved to compensate for not having the mental capacity to juggle all the potential correlates in the real world. In other words, given enough data, we wouldn’t have needed Kepler to figure out orbits. The scary part is it may figure out GUT - or some ways to harness energy - and we won't know what "it" means.
That's a very strong claim about LLMs in general or it's drawing a semantic line between LLMs and AIs that will be referring to real world stuff
As I said in the post that disteessed Bob so much
Dunno. I'm skeptical that LLMs will be sufficient but I also don't think the hallucination objection is a serious one to the LLMs in general
My guess is that we will se the real action start soon with robots wandering around modeling the world and making decisions. They will incorporate LLMs or maybe will just be considered to be extensions of LLMs. They're likely to have their autonomy as well as connecting to distributed and/or centralized systems.
These will link symbols to the real world. The reason I'm not sure they can't still count as LLMs is that humans (and AI) have a model of the real world which we actually refer to (and try to keep coherent with the real world. I's not clear philosophically how much the real world matters (or even that it exists)
This is an interesting philosophical debate but in terms of practical AI it's just whether we decide to call them something other than LLMs. Either way AIs that link symbols to what we consider the real world isn't science fiction.
This is an interesting philosophical debate but in terms of practical AI it's just whether we decide to call them something other than LLMs. Either way AIs that link symbols to what we consider the real world isn't science fiction.
I agree, it’s really just the notion of AGI that makes for good science fiction. But as I’ve said, AI/LLMs don’t have any skin in the game and that’s actually a good thing.
Last year the US added an average of just 15,000 jobs a month, very few by historic standards.
The employment slowdown has raised concern about the health of the economy, but evidence of wider deterioration is elusive.
Layoffs have remained limited, apart from some high-profile cuts at firms such as Amazon and UPS and the unemployment rate has held steady at around 4.3%. Meanwhile, the wider economy continues to grow, expanding at a robust annual pace of 4.4% in the most recent figures.
It's a puzzling, and unusual, mix.
"It's actually very hard to point to another moment in the last 25 years where you have the combination we see today," said Jed Kolko, senior fellow at the Peterson Institute for International Economics.
Last October the investment bank Goldman Sachs put out a report, which was widely cited, suggesting the US could be facing a new period of "jobless growth" thanks to the arrival of new technology and artificial intelligence (AI) in particular, allowing companies to do more with fewer workers.
Concerns about the wider implications of such a change pulsed through discussions at the World Economic Forum in Davos last month and have contributed to widespread economic anxiety in the US.
Similar story in the Uk where our dependence on service industries may make it worse.
It's all marginal and messy enough to still deny but it's also just the tip of the iceberg that we're hurtling towards.
