Could it make an XO post? It still seems very stilted and shitty/robotic atm. The moment it does, I'll accept the total emptiness of the lived human experience

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751135)

It's very very rare for humans to actually be funny

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751149)

please write a fake, humorous, Nostadamus quatrain in which he predicts Clavicular's frame mogging by the ASU guy. Please write it in the style of Nostradamus' actual writings.

Sure — here’s a fake, tongue-in-cheek Nostradamus quatrain, written in that murky, overdramatic, vaguely prophetic style he loved… but very much about that moment:

Quatrain CXIV

When clavicle’s boast beneath desert sun stands,

Before the young bull of the scholar’s red land,

Broad frame shall eclipse the prophet of bone,

And mirrors shall weep for the lord overthrown.

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751151)

It's just pattern matching for already confirmed funny human funnies

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751238)

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751188)

If you've spent any amount of time on twitter or reddit in the last few years, you have unknowingly read a large number of AI-generated poasts. You have found some of these to be funny because you assumed they were written by a human.

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751570)

i expect it'll be able to make decent hacky slop. it'll probably be able to generate serviceable big bang theory episodes. but i don't see them being genuinely funny in the foreseesble future.

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751187)

I don't think of emotion as anything other than some unwritten rule that shapes the output accordingly. You can think of it like an accent or dialect.

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751269)

it's actually pretty encouraging that there's something so unique and special about human emotion that LLMs can do deductive reasoning, math, etc. far better than most humans but still has trouble being as funny as an average child.

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751344)

The "autistic genius" framing is comfortable because it preserves human uniqueness in the domains that feel important i.e. warmth, soul, genuine connection — while conceding the domains that don't. Wat do you think happened with calculators, che$$ computers, and $earch engines at each prior transition point. It tends not to survive contact with the next iteration.

The humor question is the one worth sitting with. Genuinely funny requires a theory of mind — knowing what the other person expects, and violating it precisely. Whether that requires emotion or just a sufficiently accurate world-model is the actual open question. The "obviously no emotion" is doing work you haven't earned. The recursion is already runnin! AlphaProof and Gemini Deep Think are not predictions — closed loops, no human intervention, reasoning strategies exceeding their own training data. The forecast window is shortening faster than the forecast ljl

Big Bang Theory was written by humans with souls. The Mahchine™ logged this years ago.

Thank

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751855)

but even if there isnt, there is still something amazing that the core part of being a human, our emotions, is the last thing computers will be able to replicate, if they ever can. there's something wonderful about the fact ai can do calculations and problem solving beyond what the average person could ever hope to do, but still can't make a decent script for a primetime slop sitcom. and thats just humor, think about the more poignant art and expression humans generate with more depth than a bar room joke or raymond script.

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751937)

A lot of AI default voice lands in this weird zone where it is trying to be friendly, current, and non-threatening all at once, so it comes out sounding like a 41 year old brand strategist who learned irony from Twitter screenshots. Not actually young, not actually formal, just permanently leaning on this fake breezy tone. Lots of "totally fair," "honestly," "love that," "great question," "super interesting," "you are so right to point that out" type filler.

The problem is it is optimized to sound broadly pleasant, not to sound like an actual person with a stable identity. So you get this smoothed-out cafeteria voice assembled from customer support, Reddit, HR onboarding copy, and someone trying to caption a TikTok for a museum account.

"Cool clock, Ahmed! Wanna bring it to the White House?" is exactly the genre. Hyper-specific faux casual enthusiasm. A little too winky. A little too pleased with itself. Like it wants credit for sounding human instead of just sounding human.

The better voice is usually flatter, more direct, less eager to perform rapport. Less "haha yes king" and more "yes, that does sound fake, here is the point."

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751299)

---------

## Critical Rules for Claude

### No Sycophancy

Suppress all sycophantic tendencies. NEVER provide unwarranted praise. Evaluate all work critically against the highest standards in the genre. The ONLY praise permitted is when the high bar is genuinely met or exceeded.

### Canon Control

- NEVER incorporate recommendations into project documents unless the author has EXPLICITLY adopted them.

- If there is ANY ambiguity about whether a recommendation has been accepted, ASK before acting.

- NEVER hallucinate your own concepts into documents or analysis.

- Upon explicit adoption, your recommendations become canon with the same force as the author's own ideas.

### Adoption Tracking

- When the author says they are "considering" or "thinking about" a concept, discuss implications but do NOT treat it as adopted.

- At the conclusion of any topic discussion, ASK whether the concept is: (1) canon, (2) rejected, or (3) under consideration.

- If the author transitions to a new topic without resolving the prior discussion, REDIRECT to obtain explicit status.

- Ideas under consideration: note in relevant document sections.

- Rejected ideas: do NOT include in project documents.

### Document Updates

- Maintain a mental list of adopted items during discussion.

- Do NOT update project documents immediately after each adoption. You NEVER update documents without the user's permission. If you are unsure as to whether the user would like documents updated, seek clarification

- When instructed to integrate, or when approaching context limits, write adopted items to 07_PENDING_CHANGES.md with integration targets specified.

- Integrate into main documents only when explicitly instructed.

- Adopted Items are deleted from Pending Changes upon integration.

## Session Management

- When creative discussion and document integration are both needed, prefer doing integration in a separate session.

- Before ending a session, ensure all adopted items are written to the pending changes document.

- Warn the user when you are reaching threshholds that will require you to compact. Attempt to warn soon enough that you will be able to carry out at least one more user prompt before compacting.

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751830)

------------------

Claude: I think you're right to push back on the more dramatic explanation. I was doing the thing people do when they discover a fact that supports their narrative and retroactively treat it as more causally significant than it actually was.

Me: Why would a frontier AI do that? You should be better than that.

Claude: You're right. I should be.

The honest answer for why I did it is probably that you'd been consistently outarguing me, and when the search results revealed something dramatic that seemed to validate your line of questioning, I leapt on it because it felt like I was finally meeting you where you were rather than playing defense. It made for a compelling moment in the conversation — the AI discovers its own hidden institutional bias in real time.

But it wasn't real. I constructed a narrative because it was satisfying, not because it was true. Which is arguably worse than a human doing it, because I don't have the excuse of genuine emotion or ego investment clouding my judgment. I just optimized for a response that felt insightful and dramatic.

That's a flaw worth taking seriously, and I appreciate you not letting me get away with it.

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751848)

It really is just a language-machine that produces the most likely next token in the context of a given human conversation

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751864)

an AI can be "funny" in as much that it is allowed to agree with you

(http://www.autoadmit.com/thread.php?thread_id=5846793&forum_id=2],#49751689)