iiinova — where your digital self keeps evolving

100%. An extended family member once dropped the r-slur at a backyard BBQ at my place, and I just said "We don't use that word in this house." He stammered a bit and I just repeated, "We don't use that word," like I was talking to a toddler. I've never heard him use that word since.

2172

Eron Main @eronmain.bsky.social · 29/09/2026

Live feed from Elections BC HQ:

Frustrated man throws his hands in the air and says "I guess!"

1190

Eron Main @eronmain.bsky.social · 29/09/2026

I can't help with 1) but for 2) as a Class of '66 Gen X, I can say: a lot. No, more than that. No, MORE. (I tell my kids about my high school's smoking area and their little Gen Z heads go "boom!")

1110

Eron Main @eronmain.bsky.social · 24/09/2026

Are you suggesting that I am merely presuming that AIs are not capable of understanding, without any foundation or consideration of facts and evidence?

100

Eron Main @eronmain.bsky.social · 24/09/2026

Sorry, I'm not following. Are you saying that if I presume AI output closely resembles human output, then it is reasonable to presume the processes that produce those outputs are the same?

100

Eron Main @eronmain.bsky.social · 24/09/2026

"Turns out, evil pays better! ¯\_(ツ)_/¯"

030

Eron Main @eronmain.bsky.social · 24/09/2026

The number of times I've had to tell my mother-in-law that the AI doesn't actually know the answer, even if it sounds like it does...

130

Eron Main @eronmain.bsky.social · 24/09/2026

Part of my knee-jerk antipathy to LLMs as an everyday tool is how good they are at fooling people into believing they are actually answering a question, after careful consideration and thought, rather than outputting a series of statistically predicted responses.

340

Eron Main @eronmain.bsky.social · 24/09/2026

So my considered and non-flippant response is that I don't actually know what would change my mind, I will have to think about it. That's not to say my mind can't be changed, just that I'm not sure how. Which is a response that I do not believe an LLM would provide.

110

Eron Main @eronmain.bsky.social · 24/09/2026

Problem is, we can't actually know if that happens. I'm sure someone could program an LLM that could give me a wrong answer, and then say that it lied for some ulterior internal motive, but that could just be another string of token responses.

100

Eron Main @eronmain.bsky.social · 24/09/2026

My initial, somewhat flippant response, would be that if an AI system lied to me, deliberately, for some ulterior internal motive, it would be the most human thing I had ever seen it do.

210

Eron Main @eronmain.bsky.social · 24/09/2026

Yes! I recognize how powerful this tool is for that function. And I see that in many many cases, the output produced is a valid response to the input. But when the output is not a valid response, it is for very different reasons than why a human would produce invalid output.

140

Eron Main @eronmain.bsky.social · 24/09/2026

I didn't say they were nothing like us. You actually took exception to my use of the term "human-like". But I am happy to say they're not alive, not conscious, not sentient, and don't have feelings. I don't need them to be that way, I just believe that they are.

130

Eron Main @eronmain.bsky.social · 24/09/2026

If you're saying an alien culture could learn a lot about us from our LLMs, I completely agree. They are a cultural artifact. But they aren't the same as us.

130

Eron Main @eronmain.bsky.social · 24/09/2026

Yes, but being a product of human culture does not mean being human.

240

Eron Main @eronmain.bsky.social · 24/09/2026

It didn't think about it and decide that was the correct answer. It didn't think about it and decide to lie to you. It didn't decide anything. It crunched some numbers and produced a result.

110

Eron Main @eronmain.bsky.social · 24/09/2026

If you ask an LLM a question and it gives you an incorrect response, there is only one reason. It is because that is the output its process has produced.

120

Eron Main @eronmain.bsky.social · 24/09/2026

If you ask me a question and I give you an incorrect response, there are many possible reasons. I might be wrong. I might be lying to you. I might have some other motive to give you inaccurate information.

110

Eron Main @eronmain.bsky.social · 24/09/2026

I don't think it's vague. You seem to understand what I mean quite clearly. LLM produce output similar to the output humans produce. The fact that LLM output is similar to human output does not mean that the LLM process leading to that output is similar to the human process.

210

Eron Main @eronmain.bsky.social · 24/09/2026

Vacuum cleaners are a product of human culture. If we meet aliens, and they have vacuum cleaners, then their vacuum cleaners will be alien to us. We will have human vacuum cleaners and alien vacuum cleaners.

1230

Eron Main @eronmain.bsky.social · 24/09/2026

Right. It cannot look at a question and realize that it doesn't have the answer. It just accepts an input and produces an output that looks like an answer. It does not understand the question. So it is not answering the question. Even if the output is correct.

010

Eron Main @eronmain.bsky.social · 24/09/2026

If I ask a question and the person or thing I'm asking doesn't have the answer, the correct response is to say "I don't know," not to make shit up. If a human made shit up when it didn't know the answer, I'd stop asking them. If it is intended to function that way then it is poorly designed.

200

Eron Main @eronmain.bsky.social · 24/09/2026

Yeah. The correct word is "malfunctioned." It's like saying, "I hit the brakes, but my car hallucinated that I was trying to accelerate."

131

Eron Main @eronmain.bsky.social · 24/09/2026

The output of these systems is human-like. That does not mean the systems are human-like. It may well be unsettling that a machine produces output that looks like something a human would do. The answer is not to assume the machine is in any way human.

3750

Eron Main @eronmain.bsky.social · 24/09/2026

Talking about AI in a way that does not ascribe human behaviors, traits, or emotions to it is not "coded language," it is accurate language. It enhances understanding. The fact that a machine produces output that resembles something a human might say does not make it human.

1210

Eron Main @eronmain.bsky.social · 23/09/2026

As a scene, with all the visuals and Robin Williams' vocals, "Friend" is hard to beat. But does it really work in any other context? (Would the Will Smith version have made it half this far?) As a song, "Part of Your World" is a classic. Part of the canon.

150

Eron Main @eronmain.bsky.social · 23/09/2026

Holy crap. Surgical precision.

0120

Eron Main @eronmain.bsky.social · 19/09/2026

I hate the fact that I understand every word in this post.

010