> Sent: Monday, March 30, 2026 at 11:41 AM > From: "Jean Louis" <[email protected]> > To: [email protected] > Cc: "emacs-orgmode Mailinglist" <[email protected]> > Subject: Re: Literate LLM programming? [Re: Is org-mode accepting AI-assisted > babel ob- code updates?] > > On 2026-03-29 11:26, [email protected] wrote: > > I think this example shows pretty well where the lie is in the > > current wave of AI. It's not the "hallucinations", it is the > > fact that they are wired to "talk" to us as if they knew what > > they're doing. > > The assertion that AI systems are inherently deceptive due to their > conversational design—particularly the perception that they "know" what > they are saying—is a common but misinformed critique. This perspective > conflates the output behavior of large language models (LLMs) with > intent or truthfulness, which are attributes of human cognition, not > machine-generated text.
LLMs generate probabilistic text patterns without intent, consciousness, or knowing what they output. For the most part, they are not designed to be deceptive. But with the addendum that the majority of AI isn't built for transparency but for perceived competence. Failure rates (hallucinations, errors) are masked through confident phrasing and guardrails, prioritizing user trust over raw accuracy stats - since visible flaws erode adoption faster than hidden unreliability. What should be mentioned is that AI usually hides statistical limits from users. They're statistical tests for the uneducated - as commonly seen in medical and humanities studies - where the mathematics is not fully understood by its users. > - LLMs are statistical models trained on vast corpora of text data. > - They generate responses based on patterns in training data, not on > understanding, intent, or factual verification. > - The ability to "talk" coherently is a feature of their architecture, > not evidence of knowledge or deception. > > - The accuracy of LLM outputs is fundamentally determined by the > quality, relevance, and bias of the data used in training. > - Organizations that curate and train models bear responsibility for > data selection and curation. > - Misinformation or biased outputs stem from training data that reflects > historical, societal, or editorial biases—not from the model’s inherent > nature > > - Humans are also prone to misinformation, cognitive biases, and > propaganda—often internalizing false narratives through repeated > exposure. > - The prevalence of propaganda in media, politics, and education > demonstrates that humans are not inherently more truthful or discerning > than AI systems. > - The difference lies in transparency: humans often believe they are > reasoning objectively, while LLMs generate responses without > self-awareness. > > The reliability of LLM outputs depends entirely on how they are > deployed: > > - Unfiltered chatbots may generate plausible but false content. > - Engineering-grade applications (e.g., mine safety protocols, > geological modeling) use LLMs as assistants within verified workflows, > with outputs cross-checked against authoritative sources. > > The idea that LLMs are "lying" because they speak confidently is a > misattribution of human traits to machines. The real issue lies in how > these tools are used, not in their design. When properly programmed, > integrated, and monitored, LLMs are powerful aids—not sources of > deception. The responsibility for accuracy remains with the human > operators and data curators. > > Practical example: when handling my tasks and notes, the LLM is to show > absolutely accurate what is inside, what is next, what I have to do, as > it is using personal context. If you are asking for historical context, > it depends on training, it can be as well very accurate. > > Train one yourself to provide you the accurate information on what you > need. That is exactly what people do. You got the base model, make it > accurate on the knowledge you wish and want, that is the power of it. > > -- > Jean Louis > >
