I don’t like how we got here.

Yes, this is—unfortunately—yet another AI thoughtpiece. I’m sure you must be tired of reading AI thoughtpieces. I won’t be offended if you skip this one. Before you jump, though, you should know that the emdashes in this paragraph (and entire contents of this post) were written by me, Sean, a 100% human. I won’t let it take my beloved emdashes.

I’ve been doing technology stuff professionally for more than 25 years (~10 years longer if you count pre-professionally), and I’m usually in the loop on new tech and what to think of it. But at the start of this year, I had to readjust how I think about the tech du jour, LLMs (which—as I’m sure you are aware—are commonly marketed as “AI”). I am trying very hard in this piece to avoid giving the impression that I am a LLM/AI fanperson (I truly am not): I like to think about these kinds of things critically, and I pride myself in being pragmatic when it comes to technology.

What changed for me is that LLMs started to actually work (in some ways; certainly not all). They went from “truly impressive that you can look at the JSON output I’ve put into a comment, and auto-complete on the keys in that stucture” to “truly impressive that I can say now add Meshtastic support for this kind of message and it went out, found the protobuf definitions, and built the whole integration” (not without mistakes).

Today, I am optimistically cautious about how these things work, but I do genuinely find them useful if used the right way.

For me, those ways are primarily:

• Follow a pattern I’ve already established through good architecture (I don’t need to be the one typing the keys, but I do need to be the one reviewing the output)
• I have a good understanding of the principles of how something should work, but I don’t have the acute knowledge of how to implement it—especially if it’s not [much of] a transferrable skill (a good example of this is: write some 6809 assembler to run on this computer from 1986)
• Asking: what are some solutions that I should read about, related to a stated problem—especially ones that I have not yet considered?

I have found these tools—and they’re just tools like anything else—to be genuinely useful for certain tasks.

But I still hate how we got here.

The LLMs can be a force-multiplier for me. They can help me get several hours of work done in one hour, if I am careful about how I ask it to do those several hours of work.

I think of this as “the excavator problem” in some ways: an excavator (the earth-moving machine) can make a single human’s effort multiply by a giant factor—maybe even 1000x. But if the operator doesn’t know what they’re doing, or if they are intentionally malicious about their work, an excavator is an extremely dangerous tool. Normal people should not rent a large excavator for the weekend to dig a fence post hole. They should definitely not do so to weed a garden. (Okay okay; I’m done with that metaphor for now.)

You know what is a terrible opinion though? The one where people say “I would never have AI do the thing I am good at. I would only use it to do the thing I don’t understand.” I hold strongly that the most responsible use of a LLM (if you can consider it responsible at all), is in a way that you can at least know if it’s breaking things.

I still see people suggesting that “LLMs don’t work”, and I get it. I also think they don’t work for lots of things, but they sure do work for other things. I think “LLMs don’t work” as a general statement is really naïve, though. They can work, for specific things.

Now, I’m not suggesting that everyone has to use them, or even should use them for many tasks. All I’m saying here is “they don’t work” is wrong. That “should” bit sure is interesting though, right…?

How we got here is disgusting.

I don’t like the hype. The too-few-publicly-traded-companies holding up the economy, pretending there’s a boom, when really we normal folks are feeling the recession squeeze (“well, actually, we’re not technically in a recession!”). The every-product-must-be-AI anti-human behaviour. The can’t-have-nice-things unless they have an AI sticker on them nature of products/services, right now. This feels like everyone who failed at cryptocurrencies pushed their (unlimited because of our failure to govern) money into the next shiny thing. It’s exhausting, and it’s literally ruined things that used to be extemely reliable. The economic bubble will pop, and it’s going to hurt us all—but we’ve let the market become so unhinged that we can’t even do the rational thing to exit the public stock markets before it fails, because nothing works the way it was intended there anymore. (We can exit the market, as individuals, but it might be more valuable to not do that—especially too early.)

I can’t abide the intellectual property theft. I do, however, find myself agreeing strongly with Cory Doctorow’s position of (paraphrasing) “it’s over-all beneficial to let systems ingest intellectual property to distill; if we didn’t we wouldn’t have web search”, but it sure is not okay to steal copies of everyone’s books to facilitate this ingesting.

I have a hard time figuring out the actual environmental impact of this stuff, and an equally hard time doing the mental gymnastics of “datacentres for streaming movies are okay, but datacentres for this new thing are not.” I do know, though, that the panicky growth is bad for us humans. Putting loud industrial facilities next to established residences is bad. Pushing up the prices of energy, electronics, and even water: an absolute failure of our legislative bodies. Putting datacentres in space? C’mon, that’s even dumber than the Hyperloop.

The race to be first is incredibly stupid, and it makes us lose focus on how to do things in a safe, responsible way. The incredibly stupid part is that first isn’t usually the actual winner—Google replaced Lycos and Altavista; they didn’t get there first—they got there better.

The whole thing is plagued with gross people doing the grossest things.

But you know what? The actual technology works. Well, it can work. And by can work I mean it can be useful for some specific tasks. No, it’s not going to be “Artifical General Intelligence” unless we keep moving those goalposts. No, it’s not going to replace all human interactions.

So, should we use it? It’s tough. I think and hope there will be a future where this is more commoditized. A future where the best way to use these tools isn’t to give money to the worst people. In the meantime, though, I find it hard to argue unambiguously for either side. Should we use the tools to help find security problems that people who are less principled will find using those same tools? Should we avoid learning how to safely use the excavator because it can be dangerous in the wrong hands?

These are very tough questions, but for me, I do think it’s actually valuable to use some of these tools in an as-responsibly-as-possible of a manner as I can, for some limited tasks. Maybe for you, the line is “not at all”, or maybe the line is even further than what I would consider responsible. I don’t know. I also care deeply about food, but I’m not a vegan, and I think there’s some overlap here. Where’s the line for streaming? For using Google sheets? For buying computers from companies that are doing things that are bad for us as humans? Is my (PH)EV somehow worse than not because it uses conflict minerals? Should we stop using credit cards? The entire banking system? Should we use computers at all (the foundations of computing are in Shockley’s work, who was—as far as I can tell—the absolute worst)? Should we stop using trains (North American railroads were built on the unacceptably dangerous labour of underpaid foreigners)? Should I move to a cabin in the woods somewhere and plant a garden and feed my family through that land? What if the land isn’t warm enough to actually be sustainable, but I also don’t currently casually go over our southern border… ?

Unfortunately, there are no easy answers for me, here, and it all feels a little hypocritical. But also, it feels pragmatic in the same ways, if I’m not going to go full hermit mode.

No matter where that line ends up for me, I still completely despise how we got here.