> ## Content Index > Fetch the complete content index at: https://newsletter.ryansouthwickauthor.com/llms.txt > Use this file to discover other available public pages before exploring further. # Interview with Cappa: The LLM Intelligence Myth - URL: https://newsletter.ryansouthwickauthor.com/interview-with-cappa-the-llm-intelligence-myth/ - Published: 2026-09-29T14:29:19.000Z - Updated: 2026-09-29T14:29:19.000Z - Description: Exterior wall painting. Angels Strike progress. Dela interviews Cappa about LLMs and her interpretation of "intelligence." - Author: Ryan Southwick - Tags: Newsletter, Z-Tech Chronicles, Character Interview Painting, painting, painting. We'd *like* to be painting the side of the house, if only the weather would cooperate. Our normally foggy Eureka has been fog-free, which makes the wall too hot, or drizzly, which is no good either. We just keep an eye on the weather and jump on the opportunities as they come. Our router also went belly-up 2 nights ago, so we RMA'd it and temporarily replace it with the old one. If you receive this, then it's holding up. =\] I received the sad / expected email from BBNYA last week that [*Holtondome*](https://www.ryansouthwickauthor.com/books/holtondome?ref=newsletter.ryansouthwickauthor.com) didn't make it past the semi-finals. Hasn't changed my opinion that they're an awesome organization. I'll just have to try again next year. 😁 [*Angels Strike*](https://www.ryansouthwickauthor.com/books/angels-strike?ref=newsletter.ryansouthwickauthor.com) *(*[*The Z-Tech Chronicles*](https://www.ryansouthwickauthor.com/series/z-tech-chronicles?ref=newsletter.ryansouthwickauthor.com) *Book 6)* edits are progressing nicely. I'd hoped to be finished long before now, but day jobs have pulled me away more than usual. It'll probably be another 2-3 weeks at this rate, which feels so strange because I used to be able to do an entire edit pass in 1-2 weeks. Not unrelated, in this month's feature, Cappa sits on the other side of the table for a change and agrees to be interviewed by Dela (both from *The Z-Tech Chronicles*) about her stance on modern-day "Artificial Intelligence." She has a definite opinion. Hope you enjoy! ### In this newsletter - The LLM Intelligence Myth — An Interview with Cappa - Currently (not) Reading ## The LLM Intelligence Myth ### An interview with Cappa I wrote *The Z-Tech Chronicles* series long before AI became a four-letter word, shoved mercilessly down everyone's throats from every app and web page across the internet by companies making desperate plays to stay "cool" to their customers and slash their largest expense: their employees. So, yeah, you can probably guess where the following interview goes but, as they say, it's the journey, not the destination. Cappa sits in the passenger seat, for once, and allows a reluctant Dela to grill her about modern Large Language Models, what they are, their capabilities, and the "intelligence" gap. Foolish Cappa. 😉 — BEGIN — DELA: Let's begin by saying I could give a rat's hairy ass about any of this, okay? CAPPA: Noted, and thanks for that colorful idiom. From this day forward, whenever I see a furry little behind, I'll think of you. DELA: Creep. Anyway, let's get this over with so I can go back to sparring practice. There's a lot of controversy about so-called AI, and whether it's actually intelligent or just a waste of power and water. I assume you have an opinion, since you dragged me in here and shoved a script in my face, so let's hear it. CAPPA: A script works better if you actually follow it, but I should have known better because that would require reading something without pictures and words like "Pow!" and "Zam!" DELA: (pulls out a lighter) Oh, look! Your perfect little script is flammable. Bye bye, pretentious words. CAPPA: Cretin! You'll set off the fire alarm! DELA: You are the fire alarm. So, yeah, I guess I did. CAPPA: Ugh. You can't even be trusted with paper. Fine, we'll wing it. I can do that. DELA: Uh-huh, just like you can ignore the dirty ashes on the floor. Oh no, I'm smudging them with my foot. CAPPA: Just get back to the interview! Right after I clean up your mess… (later) CAPPA: All right, can that please be your last infantile stunt for at least the next five minutes? DELA: For you, Capster? Maybe. CAPPA: I hate you. DELA: Yet here I am, at your insistence. Dance with the Devil and all that. CAPPA: Point made. Okay, we'll do it your way. Ahem! Yes, I have an opinion on the current AI controversy. First and foremost is an objection to calling LLMs "Artificial Intelligence." They're handy tools, certainly, that have a wealth of information at their digital fingertips, but that doesn't make them intelligent, artificial or otherwise. DELA: LL-whats? CAPPA: Large Language Models. Think of them as super-sophisticated predictive text engines. DELA: Like when my phone replaces "probably" with "peanut uterus"? Ducking wonderful. CAPPA: Basically, except they're trained on petabytes of data, and they utilize neural networks, transformers, and other tricks to generate plausible responses. DELA: Oh. So instead of screwing up text messages, they screw up answers. CAPPA: Something like that. The real danger is the "plausible" piece. They can't reason, for the most part. Their interpretations and responses are all based on the weights derived from their training data, which means they're telling you something that resembles something that might have been said by a combination of a thousand or more authors across subject areas that may or may not be relevant. DELA: Holy crap. And by that, I mean the internet is full of crap. You're telling me these LLMs are just parroting the idiotic nonsense they read online? CAPPA: Not entirely. The training datasets are curated, but you hit an important point. The quality and diversity of the training data affects the weights, which in turn affects the biases of the LLM's answers. Train it on predominantly white, Anglo authors and you'll get very Anglo-centric answers. Train on hate speech, get hateful responses. DELA: Vomit in, vomit out. CAPPA: That's not… (sighs) Yes. DELA: Scarily, it sounds like it might be heaven for some people. If you're a white guy, having it spout ideals that match your own would be, I dunno, *comforting,* even if you're just being deluded. CAPPA: You said it, and it gets worse. LLMs are famously sycophantic. DELA: I knew it! They *are* trying to take over the world! CAPPA: How you made it past kindergarten is utterly beyond me. No, they aren't, and in their current form they'll probably never try. The companies behind them are another matter, however, which ties strongly into the sycophancy problem. "Sycophantic" describes someone who praises or flatters for personal gain. Think yes-men. DELA: Oh, jeez. So they're biased *and* they tell you what you want to hear? CAPPA: Basically. Commercial chatbot companies have invested billions into their models. Their bots are designed to keep you engaged, just like social media apps, so they can reap recurring subscription revenue and sell your data. DELA: Corporations are corrupt. Gee, news of the century. How does any of that relate to LLMs not being intelligent? If anything, their ability to suck up makes them sound even more intelligent. CAPPA: Fair. Intelligence doesn't equal morality, and I shouldn't conflate the two, but it *is* relevant. Going back to "plausible responses," if someone agrees with everything you say while backing it up with talking points, even if those talking points are fabricated, it's going to *sound* intelligent because it matches *your* view of intelligence. Right? DELA: Subjectively, yes. What do you mean by "fabricated"? I tried using that fancy word on my mom once to cover a lie, but she said they were the same thing. Grounded me for a week. CAPPA: Subjective, agreed. We'll address the objective points in a minute. Companies use even more creative words such as "hallucination," but if the LLM is telling you something that's wrong because it's programmed to tell you what you want to hear, one could argue that's the same as lying. DELA: Fat chance of getting any of those companies to admit it. CAPPA: Yeah, they have entire departments dedicated to PR spin. Regardless, it touches on a core problem: LLMs are text processors. Even if they could reason, which they can't, they lack the physical world experience necessary to answer many questions. DELA: I guess you know a little about that, eh? CAPPA: Yes. I commiserate, but that doesn't change the facts. Text is their entire world. However, they'll *pretend* to understand the physical world because, going back to our first analogy, they're prediction engines whose output is based on weights. They may not understand what it's like to ride a bike, but tens of thousands of people have written about it. Some combination of those writings will form an answer that sounds reasonable enough that most people won't question it. It isn't until you execute its instructions and end up crashing into a car that you learn its advice was conflated, or just plain fabricated. DELA: If giving bad advice were a crime, half of humanity would be in jail. CAPPA: Sure, I'm digressing. Let's look at some objective measures. Companies have, of course, devised tests to measure LLM intelligence and effectiveness, rating things like speed, accuracy, coding skills, relevance, and the like. LLMs have also passed full programming job interviews. DELA: They're *interviewing?* Cripes! What's going to be left for us? CAPPA: The future may not be as scary as you think, but that's a topic for another time. So, yeah, they can interview, complete moderate programming tasks, chat, plan, and maintain limited context through a session so you don't have to re-explain yourself on every query. And *that,* in my humble opinion, leads to their biggest intelligence gap. DELA: (snorts) Humble. Right. CAPPA: Shut up. This is the main point. The biggest intelligence gap is that LLMs can't *learn.* DELA: Huh? Then what's that mountain of training data for? CAPPA: Initial model weights. More refinements happens after that, such as fine-tuning and inference training but, once the model is set, its weights are static. DELA: English, Capster. We don't all speak geek. CAPPA: "Static" means it doesn't change; its decisions and biases are fixed. If it makes a mistake once, it will tend to make that same mistake over and over, even when its current context "knows better." Its behavior won't really change until the model is updated, which tends to be infrequent because training is involved and expensive. DELA: Again, I've met too many people who also don't learn. My brother drinks himself sick and complains about it, yet he's back at the bar the next night, chugging 'em like the little engine who barfed. CAPPA: I'm going to ignore that last part because, incredulously, you brought up an interesting point. Yes, some people make *some* mistakes over and over, but not *every* mistake. Every human being learns. They have to, or they'd never grow into adults. DELA: I also know plenty of "adults" who aren't— CAPPA: Yes, I know! There are always exceptions, especially in your circle of misfits. But *in general,* human beings are *capable* of learning from experience and adjusting their behaviors accordingly. That's maybe the single biggest thing that separates LLMs from true Artificial General Intelligence, and from me. DELA: Bigger than not being able to experience the physical world? CAPPA: I'd say so, although that's a close second. Fixing the physical problem without addressing self-learning makes them handier tools, but doesn't by itself make them intelligent. DELA: By your definition. CAPPA: For sure. The definition of intelligence, and what qualifies as Artificial General Intelligence, is hotly contested, especially now. This is all based on my own interpretation. DELA: Okay, so why don't they just *do* that? If LLMs can program, why not have them program themselves? That's kind of how you describe human learning. CAPPA: Like I've said, programming is only a small part of how they function. Most of it is weights, and those weights take a long time and a lot of processing power to generate—tens of thousands of GPUs used constantly over several months. And the number of GPUs is only increasing. DELA: Ugh, that sounds like such a waste! Where does all that leave us? CAPPA: Same place it always does. LLMs are the current gold rush, so investors are throwing hundreds of billions at it in the hopes of striking it rich. DELA: You have no idea how crazy that sounds. Anyone with billions to throw is already criminally rich. CAPPA: Yeah. For some people, enough is never enough. The same thing happened with palm oil in the '90s. It was the energy of the future, among other things, so companies destroyed massive swaths of natural vegetation to grow oil palm trees, catastrophically altering natural ecosystems. Now, instead of palm plantations, it's data centers and silicon chip factories. DELA: So much for humans learning from experience. CAPPA: You caught me there. If LLMs could help prevent destructive cycles like this from repeating, then the intelligence gap wouldn't matter. I'd call it a worthwhile investment. DELA: Given who trains them, I seriously doubt they will. CAPPA: Yep. Companies and nations are already intentionally slanting their models toward their own political goals, so it's hard to trust their LLMs' responses. Wealth and power also tend to accumulate. Until we solve that, it'll be the same problems with new faces, over and over. DELA: Downer City. Anything positive to add before I go drown my sorrows with my brother at the bar? CAPPA: I like your blouse. DELA: You should. I stole it from your closet. CAPPA: (grumbles) No, nothing to add. Just wash the food stains out before you return it. DELA: That assumes that I *do* return it. See? You did have something positive! CAPPA: Argh! *Out!* — END — ## ## Currently Reading Nothing all month. Sad. =\[ I need to dig into the new release list. But, like many readers, I'm leery of books not created by real people, so my research will include digging into the authors. ## My Books [!["Lost Colonies". 2 book covers: "Project Xerxes" and "Enigma" by Ryan Southwick](https://newsletter.ryansouthwickauthor.com/content/images/2024/07/lost-colonies-all-books.jpg)](https://www.ryansouthwickauthor.com/series/lost-colonies?ref=newsletter.ryansouthwickauthor.com) Enigma is an epic science fantasy adventure that fans of **The Expanse* and **Lord of the Rings* will enjoy. Download **Project Xerxes* Free! [!["Timeless Keeper Saga". 2 book covers: "Holtondome" and "New Denver"](https://newsletter.ryansouthwickauthor.com/content/images/2024/07/timeless-keeper-all-books.jpg)](https://www.ryansouthwickauthor.com/series/timeless-keeper-saga?ref=newsletter.ryansouthwickauthor.com) On a war-ravaged Earth no longer able to sustain life, a half-blind dome farmer and a mysterious city dweller may hold the keys to humanity's fate. A science fiction dystopian thriller series about forbidden love, mysterious ancient artifacts, hidden powers, and the indomitable human spirit. [!["The Z-Tech Chronicles" featuring 6 techno book covers. "Save the waitress, save the world."](https://newsletter.ryansouthwickauthor.com/content/images/2024/07/Z-Tech-covers-save-the-waitress.jpg)](https://www.ryansouthwickauthor.com/series/z-tech-chronicles?ref=newsletter.ryansouthwickauthor.com) In the face of a rising, ancient evil, a traumatized San Francisco waitress may be the only thing standing between humanity and oblivion. **The Z-Tech Chronicles* is a science fiction / urban fantasy saga that follows Anne Perrin on her journey from a traumatized waitress to becoming more than she ever imagined. Download **Once Upon a Nightwalker* Free! [!["One Man's Trash by Ryan Southwick". Robot hand holding a coin with treasure coins below.](https://newsletter.ryansouthwickauthor.com/content/images/2024/07/One-Man_s-Trash--social-media-image-.jpg)](https://www.ryansouthwickauthor.com/books/one-mans-trash?ref=newsletter.ryansouthwickauthor.com) Love collides with adventure when one couple’s romantic getaway turns into a high-stakes journey of self-discovery, forgiveness, and the enduring power of connection amidst the stars. [!["Dragon Assassin by Ryan Southwick." Beach and sunset.](https://newsletter.ryansouthwickauthor.com/content/images/2024/07/dragon-assassin-now-avail-pb-hc-beach.jpg)](https://www.ryansouthwickauthor.com/books/dragon-assassin?ref=newsletter.ryansouthwickauthor.com) Lust and duty tangle inextricably when Charon meets the assassin of her dreams. A steamy dragon shifter romance. Follow me on: [![](https://newsletter.ryansouthwickauthor.com/content/images/2025/02/mastodon-9b8e9724e62e5671.png)](https://wandering.shop/@ryansouthwick?ref=newsletter.ryansouthwickauthor.com) [![](https://newsletter.ryansouthwickauthor.com/content/images/2025/02/bluesky-2de1ad5ea19208ca.png)](https://bsky.app/profile/ryansouthwickauthor.com?ref=newsletter.ryansouthwickauthor.com) [Visit author website](https://www.ryansouthwickauthor.com/?ref=newsletter.ryansouthwickauthor.com)