But Have You Slopped Your Butthole?

Bad Vibes

Last month, I sat down with people that were looking for someone with "AI experience". I'm sorry, but what does "AI experience" even mean? A fundamental premise of AI, as specifically applied to corporate productivity, is that you prompt it using natural language. If I've asked a human to do something, surely I can ask a machine to do something. Right??

"Oh, sorry. You haven't vibecoded? Yikes."

Fuck me if we have reached the point where vibecoders are more marketable than actual coders. Heaven forbid a person has actual skill, as long as an apparently skillful prompt engineer can milk minimally viable code from an LLM's supple teat. #tokenmaxxing

To be clear, coding was not the actual scenario, but it illustrates the point.

But hey, let's gatekeep based on "AI experience". A mentality from those same asshats who are self-proclaimed prompt engineers; now there's the real talent.

In summary, what was apparently unclear to these people is that I am an expert in my field: I can miraculously divine the difference between good work product and bad work product, whether it be skillfully crafted by a human or mindlessly shat out by the LLM du jour.

As anyone who has worked with another human being would know (unclear whether that includes these AI-pilled individuals): garbage in, garbage out. Most of the battle is figuring out how to ask for what you want.

The same principle must not apply to something as rare, coveted, and fundamentally unlearnable as "AI experience".

GPT, by Nestlé™

Well, as ever, it turns out there is nothing new under the sun. In doing some old-fashioned Internet research (absolutely not looking at you, Claude you smarmy fuck) to confirm the veracity of the Nestlé breast milk controversy (an uncomfortable amount of effort given the blog entries of late, I know), I found another blog post drawing the very same parallels. Alas.

That post references "cognitive offloading," and the term I've heard is "cognitive surrender". Regardless, it sure seems like an inevitable business model for these AI providers is to get people hooked, keep them coming to the point where their critical thinking atrophies (or they never develop these skills in the first place, aka college students), and exploit these now-thoughtless LLM agents (the human kind) to convert a portion of their AI-supported salaries (or, in its current incarnation where companies are scrambling to AI ALL THE THINGS, overhead) into corporate revenue. It's like nicotine, but for the mind. Or something. Just wait until they jack up prices and pass through the real cost.

Which brings me to the story that crystalized this particular gem inside my mindhole. A high-level exec at a large company shoveled a colleague's report into an LLM (Lord knows which one) and had it splurt out a bulleted list of everything that was "wrong" with the document. Really makes me wonder what the prompt was.

But that alone isn't particularly problematic. I do see an inevitable mainstay of this particular incarnation of "AI" being distillation of, and ability search by, semantic meaning. I am not so convinced about its "reasoning" ability. See also "vibe physics" and "vibe genetics". But I digress. Who doesn't like a good hallucinogenic trip from time to time, especially one that is company-financed.

Anyway, rather than really engage with the output (wouldn't want to get a couple more mental reps in, it was almost 5 PM after all), this ambitious prompt engineer slopped it right into a reply email and hit "send". Zip zap ZOWIE. What a productive day he had!

But that wasn't enough, he also had to start the next day strong. So he sent "part two" of "his" complaints the following morning. Behold: an enterprising prompt pioneer.

Even a cursory review would show it was generated by machine (and a particularly dumb one at that). Many bullets complained about references to non-existent filings, which even the most basic search revealed do, in fact, exist. Other bullets were pedantic to the point of insanity. No human person, much less a high-level exec, would quibble about minor terminology (business "unit" vs. business "group", for example).

If this is how he chooses to discharge his fiduciary and ethical duties, why employ him at all? One could just deploy ShitGPT to man (deliberate) his inbox and achieve an exactly identical result. How long will these automatons remain employed?

"Human-Powered AI"

Fucking yuck. I heard this term a couple weeks back.

Human... Powered... Artificial intelligence? You mean like real artificial intelligence? Man, I wish.

No, what was meant was that someone turned around work product so quickly that it felt like AI. Which is something I am starting to see more often.

The apparent ease and vim with which LLM's produce convincing and engaging output makes the time outlay for conventional, thoughtful human-crafted content feel downright glacial. I am still (and likely will forever be) partial to that human spark, but a surprising amount of people seem indifferent or even biased toward the former. I want it NOW!

Part of the challenge is a lack of understanding where limitations exist, and, short of really interacting with machine-generated output, such limitations are no longer immediately evident. There is this ongoing, blind push to incorporate generative AI (or "Gen AI", as it is so barfingly called) wherever possible, regardless of whether there are real efficiencies to be had and whether accuracies can be maintained.

To date, there is no replacement for genuine human thought and reasoning. Maybe there will be someday (AGI when?), but I am currently more bearish than most. Until that happens, or until AI is at least consistent and infallible, I don't see how human tasks can be so blindly offloaded.

Frankly, I blame Siri. I think Siri's decades-long run of perpetual inadequacy kicked off an entire industry of product offerings that aren't even "good enough" and a population of people ready to gleefully receive them. Why are we now okay with things that are half-baked? Why would any self-respecting company (as if a company has values) ship a product that is so routinely inadequate. Not just Apple, I mean generally.

Even assuming generative AI, as it currently exists, is right 90% of the time (a totally made up stat), that isn't enough. Sure, it obliterates standardized tests. So did my book-smart college roommate. That doesn't mean I would trust him to opine on deeply practical or subtly nuanced matters. Nor would either be good at coding their way (there I go personifying AI, which is also insidious; topic for another time) out of the various edge cases in which I routinely find myself. Or maybe my prompts just suck, but then again so does ur mom.

Shit's Fucked

The false sense of security instilled by AI and the temptation to either engage in only a skin-deep review or forgo review altogether will lead to mental atrophy (or psychosis, LFG). Even in writing this, I have fought the temptation to loop in Claude for synonyms and grammar check. Is there even a synonym for "engage with"? Literally unknowable. And it doesn't help that search engines are no longer useful: either surfacing SEO bullshit, AI-generated websites, or an AI summary. Probably all three.

For those that are less skilled, I have no doubt generative AI will raise the skill floor. And that could be really great. In some ways, I think it might have a democratizing effect.

But I worry it will also lower the skill ceiling. People won't be willing to pay or even wait for the skilled, nuanced human take, let alone expend that effort themselves. In the extreme, what if this marks a turning point for human innovation? Will LLM's plateau (or digress) as new human-produced content becomes scarce and is drowned out by an deluge of thoughtless machine-generated garbage?

Where will we be if nearly all human expression is filtered through one of only a handful of LLM's? Maybe we will all just "create" in a sickly color of beige: inoffensive and just barely good enough, a whole world of machine-generated, minimally viable text, or images, or video (don't even get me started on AI-generated "art").

And how fucked will we be when there is an oligopoly on efficient "reasoning" and the populace is no longer skilled? Gotta pay your reasoning provider if you want to engage in "skilled" labor. Drink verification can, amirite?

Or, in the final form being relentlessly heralded by our self-fellating AI prophets, what happens when we've all been replaced by shitty machines? AGI at any cost! Can't keep consuming if the people aren't being paid. Also, "consumer" is so reductive, but that is a topic for a post that is many years overdue.

I do not have answers, and we are not in control. Corporations must conform, and we must submit. So join me in working to stave off mental atrophy while waiting for the impending economic collapse.

Or, more optimistically and (hopefully) more likely, the poppening cannot come soon enough. Get fucked, Sam.