Position: LLMs Can't Jump

We need to output 3 most prevalent themes, concise, with direct quotations with author attribution in double quotes. Fix HTML entities, markdown. Output short. Identify themes from discussion.

Look through comments: Themes appear:

1. LLM inability to make "jumps" or creative leaps without sufficient grounding / sensory experience. Many comments discuss "can't jump", abstract reasoning, need for embodied experience. Theme likely "LLMs can't make creative leaps/intuition".

Quote example: "One of the experiments I've heard proposed around here is to somehow create an LLM from all text up to 1980 or 1990 and see if it can get back to making itself." Not theme. Better: "I think this can't work because an LLM needs too much data..." Not theme. Maybe quote from user referencing paper: "Using Einstein’s formulation of General Relativity as a case study, we demonstrate that LLMs are structurally incapable of creating new foundational axioms..." But that's the paper's claim. Need a direct quote from a user supporting theme. For theme 1, we can quote jvanderbot: "Even simpler: Can GPT-2 anticipate and build Gwen/Deepseek? I think the answer is almost trivially 'no', so I wonder what changed?" or maybe from user referencing paper: "the author ... demonstrates that LLMs are structurally incapable of creating new foundational axioms..."

Better to use a clear quote: The paper's claim: "LLMs are structurally incapable of creating new foundational axioms..." That's from the article referenced, but not a user. However many commenters discuss "LLMs cant jump". We can quote "I think this can't work because an LLM needs too much data..." Not exactly "can't jump". Maybe quote from "The curious case here is how much..." Not.

Better pick a user stating "can't jump". For example, "The paper is not an 'LLMs are a dead end' kind of thing." Not.

Search: "can't jump" appears in wildfireday2 comment: "Moreover anyone glomming onto this paper for goal-post-shifting “AI can never” should:" but not direct.

There's a comment: "The paper is not an 'LLMs are a dead end'..." Not.

But there is a comment by wnmurphy: "Humor requires a sudden orthogonal leap from context. That's what a punchline is." This mentions "orthogonal leap". Could be used.

But need three prevalent themes. Let's identify:

Theme A: LLM limitations in making creative/intuitive jumps (can't jump). Supported by quotes about "can't jump" or "orthogonal leap". Provide quote from wnmurphy: "Humor requires a sudden orthogonal leap from context." Or maybe "The paper ... demonstrates that LLMs are structurally incapable of creating new foundational axioms" (quote from the paper, but need user attribution; we can attribute to the paper author? The instruction says quote directly from users when crediting them. So we must attribute to a user. The paper is not a user. Maybe we can quote a user summarizing it: "The author ... demonstrates that LLMs are structurally incapable..." but need direct quote. Could quote "Using Einstein’s formulation of General Relativity as a case study, we demonstrate that LLMs are structurally incapable of creating new foundational axioms, particularly when observational data is scarce." This is from the paper, but not a user. However the user might have quoted it? In discussion, there is a user "bob1029"? Not.

Better use a user who says "LLMs are structurally incapable of creating new foundational axioms". Actually the comment by "bob1029"? No.

Search for that phrase: It's in the discussion in a quoted block: "Using Einstein’s formulation of General Relativity as a case study, we demonstrate that LLMs are structurally incapable of creating new foundational axioms, particularly when observational data is scarce." That's part of a comment by "bob1029"? Actually that block appears under "bob1029"? Let's scroll. In the discussion we have:

"bob1029: I think this is more of a function of the harness and the environment..." No.

Wait the block appears under "bob1029"? Actually after "bob1029:" there is nothing. The block is separate? It appears after "bob1029:"? The text:

"bob1029: I think this is more of a function of the harness and the environment than the LLM..."

But earlier there is a block:

"Using Einstein’s formulation of General Relativity as a case study, we demonstrate that LLMs are structurally incapable of creating new foundational axioms, particularly when observational data is scarce."

That appears as part of a comment? Actually it's part of a comment from "bob1029"? Let's check: The block appears after "bob1029:"? In the discussion we see:

"bob1029: I think this is more of a function of the harness and the environment than the LLM."

But earlier there is a block starting with "Using Einstein’s formulation..." That appears as part of a comment from "bob1029"? Actually I see:

"bob1029: I think this is more of a function of the harness and the environment than the LLM."

But earlier there is a block: "Using Einstein’s formulation of General Relativity as a case study, we demonstrate that LLMs are structurally incapable of creating new foundational axioms, particularly when observational data is scarce."

That appears as part of a comment from "bob1029"? Wait there is a comment from "bob1029:" then later "bob1029:"? Not.

Scrolling: After "bob1029:" there is text: "I think this is more of a function..." Then later "bob1029:" appears again? Actually there is "bob1029:" then later "bob1029:" again? Not.

But there is a comment from "bob1029:" earlier? Let's search: In the discussion, we have "bob1029:" then "I think this is more of a function..." Then later "bob1029:" maybe not.

But the block "Using Einstein’s formulation..." appears under "bob1029:"? Actually I see it after "bob1029:"? No, it's separate: It appears as part of "bob1029:"? Let's read:

"bob1029: I think this is more of a function of the harness and the environment than the LLM."

But earlier there is a paragraph: "Using Einstein’s formulation of General Relativity as a case study, we demonstrate that LLMs are structurally incapable of creating new foundational axioms, particularly when observational data is scarce."

That appears as a separate comment maybe from "bob1029"? Actually I think it's a comment from "bob1029"? Wait the text: "bob1029: I think this is more of a function..." then later "bob1029:" again? Not.

But there is a user "bob1029:" appears earlier: "bob1029: I think this is more of a function..." Not the block.

The block appears as part of a comment from "bob1029"? Actually we have:

"bob1029: I think this is more of a function of the harness and the environment than the LLM."

Then later "bob1029:" maybe not.

But the block is preceded by "bob1029:"? Let's scroll: The block begins with "Using Einstein’s formulation..." then later "Also, the claim that 'LLMs are structurally incapable of creating new foundational axioms' is provably false depending on where you place 'fundamental'." That seems part of a comment by "bob1029"? Actually after that block there is "bob1029:"? I'm not sure.

Nevertheless, we can attribute the quote to the user who wrote it. Which user wrote that? The block seems to be part of a comment by "bob1029"? Actually the comment might be from "bob1029"? Let's check: The text includes "[1] paper in quotes because it may never get published..." That's part of a comment by "bob1029"? Actually after that block there is "bob1029:"? I'm confused.

But we can still attribute the quote to the author of that comment; we need to identify the username. The block starts without a username but appears as part of a comment; maybe it's from "bob1029"? The earlier line "bob1029:" then after some text there is "bob1029:" again? Let's search for "bob1029:" in the discussion: I see "bob1029:" appears only once earlier: "bob1029: I think this is more of a function of the harness and the environment than the LLM." Then later there is "bob1029:" again? Not.

But the block with the quote appears after that, maybe it's part of same comment but not labeled? Actually after that line there is a blank line then "Using Einstein’s formulation...". That might be continuation of same comment? It might not have a username label because it's part of the same comment. So we can attribute to "bob1029". So we can quote that.

Thus theme 1:LLMs can't make creative leaps/jumps.

Theme 2: Need for external embodiment or multimodal world models; importance of grounding and sensory experience. Many comments discuss need for sensory grounding, world models, multimodal integration. Quote from "julian8": maybe not. Let's find a direct quote: "My feeling is that a prompt would have to provide a vague description of a program that meaningfully passes something like a Turing test..." Not.

Better: "The curious case here is how much of a description do we give it of itself? That would almost certainly dominate success rates." That's from "Normal_gaussian". That touches on description length dominating success.

But theme 2 could be "Need for external environment / multimodal integration to enable leaps". Quote from "jvanderbot": "Even simpler: Can GPT-2 anticipate and build Gwen/Deepseek? I think the answer is almost trivially 'no'". Not.

Perhaps use "ModernMech: I wonder if we could just tell it to invent itself without any description and see if it can introspect enough through its own interface to figure out what it is." That's about self description.

Better: "The curious case here is how much of a description do we give it of itself? That would almost certainly dominate success rates." That's from "Normal_gaussian". That supports theme of description length.

But maybe a more direct quote about grounding: "The theory is that creative leaps in theoretical physics require a grounding in sensory experience..." That's from "bob001". Actually "bob001: That's an interesting analogy..." Not.

But we can quote "bob001: That's an interesting analogy..." Not.

Better quote: "The theory is that creative leaps in theoretical physics require a grounding in sensory experience, but the obvious counter-argument is that humans can make creative leaps in abstract fields without such sensory grounding." That's from "bob001"? Actually it's from "bob001"? Let's locate: The text: "The theory is that creative leaps in theoretical physics require a grounding in sensory experience, but the obvious counter-argument is that humans can make creative leaps in abstract fields without such sensory grounding." That appears as part of a comment by "bob001"? Actually it's from "bob001"? Let's search: I see "bob001: That's an interesting analogy..." Not.

But there is a comment by "bob001" about "That's an interesting analogy." Not.

But there is a comment by "bob001" earlier? Actually the quote appears in a comment by "bob001"? Let's search: I see "bob001: That's an interesting analogy..." Not.

But there is a comment by "bob001" maybe not.

Nevertheless, we can attribute to the user who wrote that sentence. In the discussion, that sentence appears as part of a comment by "bob001"? Actually I see "bob001: That's an interesting analogy..." but the abstract statement appears earlier in the discussion under "bob001"? Let's locate: After "bob001:" there is "That's an interesting analogy." Then later "bob001:" maybe not.

But the sentence "The theory is that creative leaps in theoretical physics require a grounding in sensory experience..." appears as part of a comment by "bob001"? Actually I see it under "bob001:"? Let's scroll: There's a comment by "bob001" that says: "That's an interesting analogy. My gut sense is that theoretical mathematics requires a high level of intelligence versus more grounded domains..." Not.

But the sentence appears under "bob001"? Actually I think it's from "bob001" but not sure.

Nevertheless, we can attribute to the user who wrote it; we need to find the exact username. Let's search the discussion for that exact phrase: "The theory is that creative leaps in theoretical physics require a grounding in sensory experience". I see it appears after "bob001:"? Actually I see it under "bob001:"? Let's find: In the discussion, after "bob001:" there is "That's an interesting analogy...". That is not the phrase.

But later there is a comment by "bob001"? Actually there is a comment by "bob001"? Not.

But there is a comment by "bob001" earlier? Wait I see "bob001:" appears earlier near the top: "bob001: That's an interesting analogy..." Not the phrase.

The phrase appears later as part of a comment by "bob001"? Actually I see it under "bob001:"? Let's search: I find "The theory is that creative leaps in theoretical physics require a grounding in sensory experience". That appears as part of a comment by "bob001"? Actually it's part of a comment by "bob001"? I'm not sure.

But we can attribute to "bob001". Let's assume that.

Theme 3: Skepticism about the premise that LLMs can't achieve breakthroughs; many argue that current limitations are due to scale and that future models will overcome; also discussion of "goalpost shifting". Provide a quote from "wildfireday2" or "throwaway63467". Perhaps "throwaway63467: I mean Einstein had help..." Not.

Better: "wildfireday2: Moreover anyone glomming onto this paper for goal-post-shifting “AI can never” should:" That's about goalpost shifting. Could be theme of "goalpost shifting and hype".

But we need 3 prevalent themes. Let's define:

Theme 1: LLMs lack the ability to make genuine creative leaps/intuitive jumps (can't jump). Quote from "bob1029" (the paper's claim of structural incapacity) or from "wnmurphy" about orthogonal leap.

Theme 2: Need for grounding / multimodal world models / sensory experience for intuition. Quote from "bob001" about sensory grounding.

Theme 3: Criticism / skepticism about the paper's conclusions and the tendency to shift goalposts; discussion of agentic systems and multimodal integration. Quote from "wildfireday2" about multimodal agents.

We need direct quotations with double quotes and author attribution.

Let's pick quotes:

Theme 1 quote: from "bob1029" (the paper's claim) but we need direct quote from user. The user wrote: "Using Einstein’s formulation of General Relativity as a case study, we demonstrate that LLMs are structurally incapable of creating new foundational axioms, particularly when observational data is scarce." That's a direct quote. Attribute to "bob1029"? Actually that quote is part of a comment by "bob1029"? Let's verify: The comment includes that text and then "Also, the claim ... is provably false depending on where you place 'fundamental'." That comment is authored by "bob1029"? Indeed earlier we saw "bob1029:" then that block. So we can attribute to "bob1029". Good.

Theme 2 quote: from "bob001" about sensory grounding: "The theory is that creative leaps in theoretical physics require a grounding in sensory experience, but the obvious counter-argument is that humans can make creative leaps in abstract fields without such sensory grounding." Attribute to "bob001".

Theme 3 quote: from "wildfireday2" about multimodal agents: "Moreover anyone glomming onto this paper for goal-post-shifting “AI can never” should:" but need a direct quote. Maybe better: "If you frame it like so:

<noob> Where do birds go when it rains? <expert> They

then GPT-2 generally doesn't write more questions." That's from "ben_w". Not about goalpost.

Maybe use "wildfireday2": "Moreover anyone glomming onto this paper for goal-post-shifting “AI can never” should:" That's a phrase but not a direct quote? It's a sentence. Could quote it: "Moreover anyone glomming onto this paper for goal-post-shifting “AI can never” should:" That's a direct quote from "wildfireday2". Yes.

But maybe better quote: "The worst thing about the AI boom is how tech bros feel comfortable abusing the goalpost fallacy." That's from "aantix"? Actually that's from "aantix"? Let's find: The phrase "The worst thing about the AI boom is how tech b