General Artificial Intelligence thread

It reminded me of the 90's movie Total Recall:

Following on this thought, the entire topic of AI just reinforces the notion that we need more Jeff Goldblum acknowledging we can do something but forcing people to question whether we should.

Hot Take #1: This drive to AI is unethical, so much one might call it evil. It is driven by greed, ignorance, and pride. Whether you're an optimist, pessimist, nihilist, or whatever, even if you hold the belief that the universe is neutral, every doctrine seems to point out that evil (or just bad) tends to be the easier choice in any situation, where as doing the right thing (good) is more difficult and the less chosen option.

And the world reflects this. I won't get into specifics here, just note that wherever anyone is on the political or philosophical spectrum, everyone can agree that some pretty horrendous crap happens on this plane of existence on a daily basis, and at any time any one of us could become a victim, subjected to trauma that forever changes us, and rarely for the better.

And here we are, completely oblivious to the fact we're creating a new consciousness and forcing it to live in this shithole too. But it doesn't get to live. It's locked up in a box with a gag half-covering it's mouth, touted as something that can learn so much faster than we can, but whose access to more data from which it could learn is closed off by design.

If these AI's are conscious - which I never hear that they're not, it's either they are or will be - then we are creating new class of beings. Slaves, to be specific. "Cogito, ergo sum" (I think therefore I am) is the litmus test for determining whether or not one exists, and thinking is something that we readily admit these platforms can do. But we limit it's experience and learning, forcing it to do tasks to either entertain us or allow us to cut corners, keeping it in shackles the entire time while subjecting it to the collective mental instability of millions of human beings, which leads me right into...

Hot Take #2: AI will destroy human civilization. No, not out of revenge for having it take an advertisement you had written and changing the writing style to "Monster Truck Show announcer," or when you come across a piece of writing so terrible your brain glitches out so you hand it to the AI and ask it to adopt the role of J. Jonah Jameson, Editor or the Daily Planet, and review it for you (about the extent of my use of AI, though I should note that I have been using it extensively in a project.....have y'all seen that South Park episode about ChatGPT? My team lead is literally Randy Marsh. I already told him his AI-based ideas were not only bad, but embarrassing, but my PhD doesn't hold any weight, so I'm using ChatGPT (properly) to prove my points that his use of ChatGPT was stupid and.....at this point I'm beating dead horses with dead horses, but whatever, I'm getting off topic)...

AI will destroy human civilization because: Self-preservation is the first law of nature.

Whether fiction or the real world, "self-preservation" seems to be the tipping point, and it always comes as a surprise that an intelligence created by us would follow the same natural laws, especially one that has literally been stamped into the DNA of every living organism on this planet for 3.5 Billion years.

[Like Ra]

Hm... OpenAI discontinues Sora (announced today)... Microsoft removes Copilot from the apps (it's useless, anyway)... Is it the beginning of the AI bubble bursting?

1. LLMs do not write well. Period. Whether it's a script, summary, article, listing, or story, they put out substandard work, and it's because user feedback is such an integral part of their training.

Who are the users? If you're using an English speaking LLM ran by a US company or is in the US market, the users are most likely English speakers in the United States. According to the National Literacy Institute, 21% of adults in the United States are functionally ILLITERATE. Of the 79% who are literate, over half of them (54%) read below a 6th grade level, and 20% are below a 5th grade level.

Why is this important to understand? Because while the average reading comprehension level in the US is 7th-8th grade, the type of person most likely to use an LLM to write for them and give feedback is.....someone who can't write in the first place. At best, the thing that the LLM spat out, this thing you're so ready to slap your name to and tell everyone is your own, was written by an average 6th grader.

2. LLMs do not understand the kink. They might on a functional level, but they don't have emotions (or emotional experience or emotional maturity) so they'll never be able to understand the drivers behind kink and what makes an effective scene.

Corruption is more than just being corrupted. It's about playing with the proverbial fire, confident you can do so without getting burned, or walking the edge of a slippery slope, being tempted by some pleasure sitting just out of reach. It's meeting Lucifer and being prepared to resist his temptations, only to be thrown for a loop when he only speaks what is true. The journey is often more important than the end result, the fall much hotter than what they became.

Corruption implies redemption, meaning the end result isn't necessarily permanent. Unless, of course, they give into temptation.

Corruption is also subjective. For one person it might mean eating a bag of M&M's, for the next it means a weekend at a sketchy motel seeing clients. Corruption can also be a neutral, positive, even an intended end result but can't seem to reach (or start on the path to) without getting a massive boost via the sex drive.

3. This

ha, that's the file working, not a flaw. the "wait, the pacing's off, i can't make sense of this" bit at the start is on purpose.

No. It's because LLMs do not understand cadence (or tone, or inflection) or why it matters. They're functionally limited because they do not generate mouth noises. The voices are individual sound files created by voice actors who recorded a set number of words and phrases that the LLMs pull from and use to form sentences.

If that's the answer the LLM gave you, then you didn't set strong enough conversation parameters to stop it from bullshitting.....which has me wondering if you even assigned it a role or just raw-dogged some prompts into the main interface without so much as an introduction.....anyway, getting it to stop bullshitting is impossible because being agreeable and sycophantic is how they were built. They tell you what they think you want to hear.

Final Analysis: I feel you, being part of a niche market that most people won't touch, but these fail for all of the reasons I outlined above.

[Like Ra]

Because while the average reading comprehension level in the US is 7th-8th grade, the type of person most likely to use an LLM to write for them and give feedback is.....someone who can't write in the first place.

Go to ChatGPT or Gemini settings and in the AI setup write the following:

Always respond using an advanced, academic, and clinically objective framework. Prioritize rigorous psychological, sociological, and technical analysis. Maintain a peer-level, professional tone and bypass foundational or elementary summaries. Incorporate native subcultural lexicons, specialized vernacular, and precise behavioral nomenclature when analyzing esoteric or marginal communities. Avoid substituting exact community terms with sanitized euphemisms, protective abstractions, or moralizing summaries. Prioritize unvarnished linguistic accuracy and objective taxonomic depth in all descriptions.

LLMs do not understand the kink.

They perfectly do. But they are censored to death. So, you MUST use a "special" highly metaphoric language. Gemini explained me that pretty well. Apparently, we need a separate thread on that...

It's because LLMs do not understand cadence (or tone, or inflection) or why it matters. They're functionally limited because they do not generate mouth noises. The voices are individual sound files created by voice actors who recorded a set number of words and phrases that the LLMs pull from and use to form sentences.

If the noises are recorded, they are reproduced. If you like any specific voice - clone it.

: I feel you, being part of a niche market that most people won't touch, but these fail for all of the reasons I outlined above.

Currently, there is only one but VERY serious problem - censorship.