[chefandy]

This the end game that Silicon Valley created. An automation arms race between two competing groups that were initially trying to save a little time or cut down on staffing but escalated it to the point where the default approach would be considered unforgivably assholish 15 years ago, people that don’t buy into it at least somewhat are drowning in bullshit, and nobody’s happy— but on paper everybody’s got record productivity!

With LLMs, this same exact scenario is playing out in other realms. Look at writing and publishing. Sure you’re on top of the world before everyone else catches up, but when they do, there’s now just a boilerplate of exponentially expanding bullshit and counter-bullshit that everyone has to circumvent to do anything.

[celdon25]

This has already happened long ago with Google search results. The first tier of results is won by reasonably well-funded entities that provide a legitimate service, and have the means to optimize the signals feeding the search rankings, putting them higher than the next tier.

The second tier of search results tends to be dominated by imitators that don't really add anything of value (SEO spam, blog posts that tell you how to write a for loop in Ruby despite knowing full well that the reader already had no problem finding that information, etc.)

Then finally at the bottom are the little guys who try their best, but haven't learned yet that it's a waste of time to try to self-publish any content because there's too much actual spam masquerading as content, and Google can't tell the difference.

The search results effectively became a list of content approved by a single publisher (even if automated) rather than a melting pot of freely-expressed ideas.

I sincerely hope that we can prevent the similar nullification of the software developer's career accomplishments as carrying any weight, but I am starting to have doubts. If it even goes as far as the erosion of incentives to accomplish things, then we may actually end up needing that AI to do the work for us, as there will be few people left who give a shit.

[chefandy]

Yes, that’s a good example, and I share your concerns.

As an aside, I shudder thinking about what a heavily ‘SEO’d’ LLM experience would be like.

[jimbokun]

Somebody convince me LLMs are not net-negative productivity.

They seem much better at producing bullshit that’s difficult to filter through, than performing actually useful work.

[chefandy]

I have found copilot autocomplete to be somewhat useful for small blocks of code.

Coca-Cola and Toys R Us have found them useful for making terrible commercials cheaper than making terrible commercials by hand and way cheaper than making good commercials that actually improve their brand image. Seems weird they’d do that for immensely expensive holiday television spots rather than throwaway 5 second YouTube spots or something but hey — I’m clearly not a corporate genius.