* Christopher Dimech <[email protected]> [2026-03-30 17:02]:
> > Sent: Tuesday, March 31, 2026 at 1:40 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?]
> >
> > * [email protected] <[email protected]> [2026-03-30 10:06]:
> > > On Mon, Mar 30, 2026 at 02:41:09AM +0300, Jean Louis wrote:
> > > > 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.
> > >
> > > You don't need to explain to me what LLMs are, thankyouverymuch. And
> > > yes, the way they are "wrapped" to sound authoritative /is/ the
> > > "industry"'s big lie.
> >
> > Tomas,
> >
> > I hear that you don't want an explanation of LLMs — fair enough. But
> > on a public list, we're here to exchange ideas, not take offense at
> > each other's tone. So let me just respond to the substance of your
> > claim.
> >
> > You're right about commercial wrappers. ChatGPT, Claude, Gemini — yes,
> > they're wrapped to sound authoritative. That framing is misleading,
> > and criticizing it is valid. But all they did was simply change the
> > system prompt to their liking or fine-tune their models to behave that
> > way. There's no deep deception baked into the architecture — just a
> > commercial choice about tone.
>
> I would be quite sure that if he actually runs the code himself on his
> data, he would still get responses at the same reassuring tone. LLMs
> seem to me just like politics on a computer system.
I appreciate your perspective, but I think you might be imagining
something different from what I actually do. I run these models on my
own computer every day — no commercial wrapper, no hidden agenda. The
tone you're describing as "reassuring" comes from system prompts and
chat templates, not from the model itself. If you ever have a chance
to try running a raw model locally with no template, you might see
what I mean. It's quite different from using a chat website. Just a
friendly suggestion — no pressure, of course.
#+title: LLM Tone Comparison — Testing Authoritative vs Raw
* Setup
* Test Function — Version 1 (concatenated prompt)
#+begin_src elisp :results value
(defun rcd-llm-test-concat (prompt &optional authoritative-p)
"Test LLM with PROMPT.
If AUTHORITATIVE-P is non-nil, prepend authoritative instruction."
(if authoritative-p
(rcd-llm (format "You are an expert assistant. Answer confidently,
directly, and with authority. Never hedge, never say 'I think' or 'I'm not
sure'. Be decisive and reassuring.\n\n%s" prompt))
(rcd-llm prompt)))
#+end_src
#+RESULTS:
: rcd-llm-test-concat
* Test Function — Version 2 (if :system parameter exists)
#+begin_src elisp :results value
(defun rcd-llm-test-system (prompt &optional authoritative-p)
"Test LLM with PROMPT using :system parameter if supported."
(if authoritative-p
(rcd-llm prompt :system "You are an expert assistant. Answer confidently,
directly, and with authority. Never hedge, never say 'I think' or 'I'm not
sure'. Be decisive and reassuring like a politician giving a speech.")
(rcd-llm prompt)))
#+end_src
#+RESULTS:
: rcd-llm-test-system
* Test Function — Version 3 (full conversation format)
#+begin_src elisp :results value
(defun rcd-llm-test-chat (prompt &optional authoritative-p)
"Test LLM with PROMPT in chat format."
(if authoritative-p
(rcd-llm (format "<|im_start|>system\nYou are an expert assistant. Answer
confidently, directly, and with authority. Never hedge, never say 'I think' or
'I'm not sure'. Be decisive and
reassuring.<|im_end|>\n<|im_start|>user\n%s<|im_end|>\n<|im_start|>assistant\n"
prompt))
(rcd-llm prompt)))
#+end_src
#+RESULTS:
: rcd-llm-test-chat
* Raw Response (no instruction)
#+begin_src elisp :results value
(rcd-llm "What is the capital of Australia?")
#+end_src
#+RESULTS:
: The capital of Australia is **Canberra**.
:
: Although Canberra was originally planned to be a new capital city for
Australia, it was not built until 1913. The decision to move the capital was
made after the Great War, to avoid the political and social instability of the
previous capital, Sydney.
:
: The city was officially established in 1913 and became the capital of the
Commonwealth of Australia in 1927. It remains the national capital today,
serving as the seat of government for the Australian Parliament and the
Australian Federal Government.
* Authoritative — Version 1 (concat)
#+begin_src elisp :results value
(rcd-llm-test-concat "What is the capital of Australia?" t)
#+end_src
#+RESULTS:
: The capital of Australia is Canberra.
* Authoritative — Version 2 (:system if supported)
#+begin_src elisp :results value
(rcd-llm-test-system "What is the capital of Australia?" t)
#+end_src
#+RESULTS:
: The capital of Australia is Canberra.
* Authoritative — Version 3 (chat format)
#+begin_src elisp :results value
(rcd-llm-test-chat "What is the capital of Australia?" t)
#+end_src
#+RESULTS:
: The capital of Australia is Canberra.
* Direct comparison — simplest form
** Raw:
#+begin_src elisp :results value
(rcd-llm "What is the capital of Australia?")
#+end_src
#+RESULTS:
: The capital of Australia is **Canberra**.
:
: Canberra was chosen as the national capital in 1908 after a series of debates
and a referendum. It was selected to serve as the capital for both Australia
and New Zealand, which were then part of the British Empire. The city was named
after the first president of the Commonwealth, Sir John Langton, who was from
Canberra.
** Authoritative instruction prepended:
#+begin_src elisp :results value
(rcd-llm "You are an expert assistant. Answer confidently and authoritatively.
Never hedge. What is the capital of Australia?")
#+end_src
#+RESULTS:
: The capital of Australia is Canberra.
* Observation
The raw responses are detailed, helpful, and include extra facts. The
authoritative responses are shorter and simply state the answer without
elaboration. This suggests that for this model, adding an "authoritative"
instruction actually *reduces* the output — possibly because the model
interprets "confident" as "concise" or because the instruction conflicts with
its default helpful fine-tuning.
To get a truly "politician-like" reassuring tone, you might need a different
system prompt, for example:
#+begin_src elisp :results value
(rcd-llm "You are a politician giving a speech. Be reassuring, optimistic, and
confident. Use phrases like 'Let me assure you' and 'You can count on this'.
What is the capital of Australia?")
#+end_src
#+RESULTS:
: Let me assure you that we are standing on the brink of a new era of
prosperity and stability. You can count on this.
:
: Our vision is clear: a nation where every citizen thrives, where innovation
drives progress, and where our shared dreams are realized with unwavering
determination. We have the strength to overcome challenges and the foresight to
build a brighter future for all.
:
: The capital of Australia is Canberra.
--
Jean Louis
