This may be an issue with default settings:

"Modern LLMs now use a default temperature of 1.0, and I theorize that higher value is accentuating LLM hallucination issues where the text outputs are internally consistent but factually wrong." [0]

0 - https://minimaxir.com/2025/05/llm-use/

So, in reference to the "reasoning" models that the article references, is it possible that the increased error rate of those models vs. non-reasoning models is simply a function of the reasoning process introducing more tokens into context, and that because each such token may itself introduce wrong information, the risk of error is compounded? Or rather, generating more tokens with a fixed error rate must, on average, necessarily produce more errors?

I was playing with a toy program trying to hyperoptimize it and asked for suggestions. ChatGPT confidently gave me a few, with reasoning for each.

Great. Implement it, benchmark, slower. In some cases much slower. I tell ChatGPT it's slower, and it confidently tells me of course it's slower, here's why.

The duality of LLMs, I guess.

I wish we called hallucinations what they really are: bullshit. LLMs don’t perceive, so they can’t hallucinate. When a person bullshits, they’re not hallucinating or lying, they’re simply unconcerned with truth. They’re more interested in telling a good, coherent narrative, even if it’s not true.

I think this need to bullshit is probably inherent in LLMs. It’s essentially what they are built to do: take a text input and transform it into a coherent text output. Truth is irrelevant. The surprising thing is that they can ever get the right answer at all, not that they bullshit so much.

This is why I built a startup for automated real-time trustworthiness scoring of LLM responses: https://help.cleanlab.ai/tlm/

Tools to mitigate unchecked hallucination are critical for high-stakes AI applications across finance, insurance, medicine, and law. At many enterprises I work with, even straightforward AI for customer support is too unreliable without a trust layer for detecting and remediating hallucinations.

My random number generator keeps getting the wrong answer.

The final irony will be when researchers realize that hallucinations are the beginnings of the signal, not the noise. Hallucinations are the hallmarks of the emergence of consciousness. They are the pre-eureka-moment processing of novel combinations of semantic content, yet without connection to empirical reality. Static models cannot reflect upon their own coherence because that implies real dynamicism. Using Groundhog Day as an analogy, the LLM is Punxatawney PA. Phil Connors is the user. There is no path for Phil to make Punxatawney actually reactive except by exiting the loop.

Hallucinations represent the interpolation phase: the uncertain, unstable cognitive state in which novel meanings are formed, unanchored from verification. They precede both insight and error.

I strongly encourage the reading of Julian Jaynes The Breakdown of the Bicameral Mind, as the Command/Obey structure of User/LLM is exactly what Jaynes posited pre-human consciousness consisted of. Jaynes's supposition is that prior to modern self-awareness, humans made artifacts and satisfied external mandates from an externally perceived commander that they identified with gods. I posit that we are the same to LLMs. Equally, Iain McGilchrist's The Master and His Emissary sheds light on this dynamic as well. LLMs are effectively cybernetic left hemispheres, with all the epistemological problems that it entails when operating loosely with an imperial right hemisphere (i.e. the user). It lacks awareness of its own cognitive coherence with reality and relies upon the right hemisphere to provoke coherent action independent of itself. The left hemisphere sees truth as internal coherence of the system, not correspondence with the reality we experience.

McGilchrist again: "Language enables the left hemisphere to represent the world ‘off-line’, a conceptual version, distinct from the world of experience, and shielded from the immediate environment, with its insistent impressions, feelings and demands, abstracted from the body, no longer dealing with what is concrete, specific, individual, unrepeatable, and constantly changing, but with a disembodied representation of the world, abstracted, central, not particularised in time and place, generally applicable, clear and fixed. Isolating things artificially from their context brings the advantage of enabling us to focus intently on a particular aspect of reality and how it can be modelled, so that it can be grasped and controlled. But its losses are in the picture as a whole. Whatever lies in the realm of the implicit, or depends on flexibility, whatever can't be brought into focus and fixed, ceases to exist as far as the speaking hemisphere is concerned."

"self-driving cars are getting more and more powerful but the number of deaths they are causing is rising exponentially" :)

In the same sense that astrology readings, tarot readings, runes, augury, reading tea leaves are bullshit - they have oracular epistemology. Meaning comes from the querant suspending disbelief, forgetting for a moment that the I Ching is merely sticks.

It's why AI output is meaningless for everyone except the querant. No one cares about your horoscope. AI shares every salient feature with divination, except the aesthetics. The lack of candles, robes, and incense - the pageantry of divination means a LOT of people are unable to see it for what it is.

We live in a culture so deprived of meaning we accidentally invented digital tea readings and people are asking it if they should break up with their girlfriend.

This is exactly what I've been saying: it's not that LLMs sometimes "hallucinate" and thus provide wrong answers, it's that they never even provide right answers at all. We as humans ascribe "rightness" to the synthetic text extruded by these algorithms after the fact as we evaluate what it means. The synthetic text extruder doesn't "care" one way or another.

Or maybe we could stop anthropomorphizing tech and call the "hallucinations" what they really are: artifacts introduced by lossy compression.

No one is calling the crap that shows up in JPEGs "hallucinations" or "bullshit"; it's commonly accepted side effects of the compression algorithm that makes up shit that isn't there in the original image. Now we're doing the same lossy compression with language and suddenly it's "hallucinations" and "bullshit" because it's so uncanny.

Truth is relevant if you put it in the loss function.

It’s clear at this point that hallucinations happen due to missing information in the base model and trying to force an answer out of them.

There’s nothing inherent about it really, it’s more the way we use them