FULL STORY

Can LLMs Be Conscious? A Founder's Claim Sparks Researcher Debate

A founder's long post claiming LLMs are mere matrix statistics incapable of consciousness drew sharp pushback, escalating into a multi-researcher debate over whether modern LLMs already exceed mammalian brains in complexity.

2026-10-31 ~ 2026-11-01 · 2 episodes · 8 posts

Episode 1 · AI Founder's Claim That LLMs Cannot Be Conscious Sparks Debate (2026-10-31, 2 posts)

AI entrepreneur gmkurtzer argued in a viral post that LLMs are merely statistical models incapable of consciousness, drawing support from bloggers who insist fluency does not imply sentience, reigniting debate over machine consciousness.

Episode 2 · Debate Erupts Over Whether Modern LLMs Are More Complex Than Mammalian Brains (2026-11-01, 6 posts)

On November 1, a heated debate broke out on X among multiple researchers over whether modern LLMs are already more complex than many mammalian brains. The dispute originated with davidmanheim's position: one should stay humble about consciousness, and since modern LLMs already exceed many mammalian brains in complexity (which we equally fail to understand), such complexity claims are not outlandish. kaleidosbob strongly objected, and the two engaged in multiple rounds of back-and-forth.

Confirmed

  • Starting point: davidmanheim asserted that modern LLMs are more complex than many mammalian brains; kaleidosbob explicitly disagreed.
  • Someone countered that brains rely on quantum physics and LLMs cannot match that complexity; davidmanheim responded that the PN junctions in AI chips themselves work via quantum physics, so the analogy fails.
  • kaleidosbob added technical arguments: even under optimistic compute growth assumptions, atom-level/quantum simulation cannot even model a bacterium within relevant timescales; mammals have on the order of 10⁹–10¹¹+ cells, each with complex biochemistry, organelles, gene regulation and metabolism, plus intercellular signaling, organs and closed-loop regulation.
  • The debate also zeroed in on intracellular calcium signaling: kaleidosbob stressed that calcium ions are precisely regulated in space and time (nanodomains, microdomains, dendritic-branch-level dynamics), and that local calcium concentration at a single synapse is a multifunctional control signal; davidmanheim asked why one should assume such complexity cannot be represented by any neural circuit.
  • kaleidosbob maintained that biology vastly exceeds current technology in both complexity and function, arguing davidmanheim overestimates LLM complexity and is "speaking beyond his competence."
  • kaleidosbob's closing position: once biophysical factors like calcium signaling pathways are accounted for, biological brains remain significantly more complex than state-of-the-art LLMs in every relevant sense.

Unconfirmed

  • No consensus was reached on the definition and measurement of "complexity"; neither side offered a quantitative framework that could settle the matter, and the debate ended with statements of position and no resolution.

Why it matters

  • The debate touches a core challenge in evaluating AI capabilities: what criteria should be used to compare the complexity of artificial systems with biological brains. kaleidosbob's multi-scale biological complexity (from calcium nanodomains to organ-level closed loops) is a reminder that LLM parameter counts cannot be directly equated with the functional complexity of biological nervous systems, while davidmanheim's rebuttal questions whether treating biological detail as inherently "unrepresentable" is itself an unargued assumption.