Chapotraphouse

Expenses are the only language management speaks so speak their language and make it expensive

"...Learn the use of expenses!"

That's the fucking thing, these assholes are now monitoring your usage. Wasteful behavior will be punished (using most expensive models for trivial tasks) as much as not using the fucking thing. I think bad prompting the cheap shit often will be most wasteful.

I've still never once been reprimanded for using too many tokens but it just makes me feel bad like there goes yet another years worth of electricity.

It's basically a large-scale "train your replacement" but your replacement is 12 drunk children

Ed Zitron straddles the fine line of soberly analyzing all of this while being absolutely clear he thinks it’s all a pile of bullshit, it’s nice

Customers who purchased this item also enjoyed The Ugly Truckling: The Story of My Cybertruck

When a dad brings home a Cybertruck, his daughter is less than impressed. With its unconventional and futuristic appearance, the truck seems like an ugly duckling in a world of familiar cars. But as the little girl begins to learn more about the Cybertruck's capabilities and its potential to help the world, she begins to see it in a whole new light.

He's gonna write it or he's gonna generate a book about it?

My manager on one of the meetings:

"You know guys, I won't beat around the bush anymore with «asking». Let me get to you straight. You have to use these tools, because one, two years from now, you'll just be left behind."

It doesn't get more excplicit than that.

grok, please summarize what the requirements at OP's office are

LLMs? Generating spam lol.

The programming stuff is waaaay overblown. It's less accurate than copy/pasting shit off Stack Overflow, the thing we used to say was the sign of a bad programmer. Everyone using it is miserable, and nobody using it has actually delivered the results they keep saying they're getting (it's been 6 months since you all said you're 100x more effective, where's the 50 years worth of new software, AI bros?)

There are occasionally these huge bursts of things AIs do, but they're all predicated on having a huge budget dedicated to finding reasons why a huge AI budget is a good idea. There was a wave of security vulnerabilities being found by AI at the start of the year, but there are existing security scan tools that are just not being used If you spent the same amount of money scanning everything on github with AFL-Fuzz, you'd find billions of security bugs. There's now a wave of math proofs being generated now, but AI companies are spending millions of dollars on this. For a million dollars you can hire a small team of math postdocs for two years. Heck, for a million dollars you can fund a permanent grant.

LLM? Not so much IMO. But neural nets and deep learning are genuinely cool and useful for things like image recognition, speech recognition, etc. As an aide to a doctor reviewing a colonoscopy scope, sure. But instead they will treat it as a replacement that the doctor rubber stamps. Due to the way we will use it, they are dangerous, even if the tech itself can be helpful.

Theres still the issue of electricity, of the human and environmental costs, etc.

A good metric: if correctness is unimportant and verification is easy, then it is not an inappropriate tool (that doesnt mean its necessarily a good tool for a given task though). However if correctness is important and verification is difficult, then it is not appropriate. Basically, dont use it to summarize that many-page report, because its nondeterministic and you'll have to read the whole thing anyway to see if the LLM actually told you all relevant portions and didn't generate things that aren't in the report. If you want a summary, thats literally what the abstract and conclusion parts of a report are for. Dont use it to write fresh code for you, cause you'll have to read through all of it, build a mental model, and understand the problem space inside and out to actually review it well, and at that point you might as well have written the code.

LLMs can translate frustration into soft corporate speak and then you don't have to think too much about how you're going to tell your boss they're the worst

Prototyping software.

If you're trying to cobble together a shitty proof of concept, and you don't care about whether it's efficient, scalable, maintainable, well-documented, or secure... then yeah, an LLM will help you do that.

But then when it comes time to actually building the real thing, you're better off using an actual team of actual humans to make it actually functional.

making your manager shut the fuck up

When I was on Xiaohongshu before they introduced the translation features, people would make fun of you for using non-LLM translation software because they could tell the difference. Either that or I would get, “Wow, your mandarin is so good!” which is the Chinese equivalent of “You are noticeably not fluent”. So it prevents bullying, but only in extremely specific circumstances.

Generating security incident reports

That shits boring

Sometimes it can write good code, sometimes it doesn't. Sometimes it summarizes some info for you well, sometimes it makes shit up. It's a mixed bag overall, probably not worth the hype and the resources but here we are anyway.

Freeform roleplay in the vein of AI Dungeon

If you're trying to search for something but don't quite have the right words to describe it, AI can help.

• summarizing large bodies of text to orient yourself in what is or isn't there. Obviously you need to open the actual documentation to check the facts, but the ai should give you links to open up.
• grammar explanations for language learning.
• it can be somewhat effective at searching across something like a ticketing or a work order system. Previously you would need to come up with a query, but sometimes the Metadata is not captured properly or at all. The llm can process the notes/comments and return a useful result.
• I've used it to draft documentation of my code, but i always have to edit it. I like having the formatting consistent at least.
• I'm not a programmer by trade, so sometimes I will have it review something I wrote and suggest improvements or a more standard way of writing it.
• slang explanations for language learning, with context.

I dont ever really use it to "do" anything on my behalf, but generally to "tell" me things. I could see myself telling a like a canva Ai to "color the 100 stars on this poster red" or something like that so I don't have to click around and change them, and I can instantly verify it was done correctly.

Is this worth the resource cost? Probably? I don't use it all the time, or even every day. Most of this could probably be handled on a local Ai instance instead of something datacenter level. I do wish to be charged the proper amount for what I use so I can make that evaluation, but that's still a bit obfuscated from me.

Certain Mathematical models seem to work well. And I've heard that some decent programmers like to use it to get together code they would have spent hours on GitHub looking for to copy and paste.

And if it wasn't so wasteful, I like the idea of using generative AI for getting visuals for (PRIVATE) TTRPGs. When I just want my group to have a better understanding of what my BBG, "Lord Gargle Deeznuts" looks like. And that's only if I can generate something decent for free. If I have to pay, it would be better to pay a real human to do the labor.

Other than than those very specific cases, I really don't believe there is any commercial use for current LLM. Everything is so completely unethical IMO, that even the grey area of generative AI that I mentioned above immediately become unethical the moment a human profits off it.

They would probably be a good replacement for CEOs and similar executive types.

Only way to stay sane in this industry anymore.