I feel like this is kind of missing the point that companies are mainly a group of humans and their roles and responsibilities matter to them emotionally. Managing those expectations and feelings can only be done by other humans that feel empathy (good managers) and abstracting such relationships onto something that can be "versioned, queried, tested, and automatically verified" might create a shitty soulless place to work.
Yes - the author's observations are not wrong. Companies, on paper, are logical. However, as one professor told us in college "All companies are perfect until you introduce the humans".
Humans are messy. Humans work outside of whatever system you create. You can codify all your things all you want, it simply will not capture the operational complexity of a business run by humans.
The problem needs to be flipped on its head. LLMs give us the capacity to do just that. It's far more accurate to analyze what the humans are doing, note deviations and follow up on those where regulatory compliance is required. This captures both written processes as well as their practical implementations.
Exactly this and its a blind spot in the article. LLM's / orchestrated specialist agents can query SOP's, policy docs, compliance docs. Having humans build these artifacts in code isn't really needed at this point. Maybe if there is an interim format LLM's can use that save tokens / time / etc. Wouldn't assume that looks exactly like exactly what human coders would have used the past though.
Anything can be used for the good or for the bad. Defining how the organization is structured and how it operates usually is usually not about how people really do their actual work -- unless there are safety etc. regulations that must be met. Many enterprises are in constant chaos, which stresses people out. Adding some structure to it helps to alleviate that stress. For example, if there is a good template to document something, you don't have to start from the scratch. Of course, you could also go all in automate all your "management", in order to avoid talking with your employees. I don't think that will end well.
Yes, but also literally AI slop. Stock characters auto-puppeted to a text script with text-to-voice run across it. The artist's input was writing the text and signing off on the final auto-generated product. The reason the two characters in the video are weird superheroes is those were the available stock characters in the XtraNormal.com service he used.
After all, the creator didn't want to be an professinal full-time animator, he just wanted to animate three minutes.
It is only "literally AI slop" if you widen the definition to include anything made using computers. That is not an honest take on what content made with AI is.
The original author chose those assets and that background, other people made those assets on the first place and had to take a ton of tiny creative choices that changed the final thing and help transmit ideas and feelings (of uncanniness, vulgarity, surrealism, whatever).
Anyone can tell the difference between one and the other.
> Yes, but also literally AI slop. Stock characters auto-puppeted to a text script with text-to-voice run across it.
Maybe I'm wrong but I would say that the slop feeling of the original version was a deliberate decision and part of its charm. That kind of video and voices were a meme back then, IIRC.
As is the slop feeling of the current, no doubt. I don't think that changes it being "slop" in the sense of low-effort and mostly generated by machine learning (rather than hand-edited like a machinima)
Most videogame cutscenes (well, for AAA games; I'm going to ignore the wide, worthy, and growing ecosystem of "I had an idea and made it in Unity with purchased assets") use bespoke model and texture assets, motion-capture human animation, and voice acting.
None of that is what an Xtranormal auto-puppet show is about.
Motion capture in bigger budget games only started to become common around the Xbox 360/PS3 generation. Go back to prior generations and it was a lot of "2 models facing each other and their mouths wiggle up and down while audio files play"
The difference being that someone actually had to record said audiofiles, and animate the mouths wiggling. The "slop"-ness is defined by the inputs, not the outputs.
Feeding text into a machine-learning based app that creates low-effort audio and visuals? Yeah, that's always been slop. I don't use "slop" in a particularly negative sense here, it is what it is
I don't think it is, actually? "Friend-slop" is a super-popular genre of video game right now (for example, indie smash-hit PEAK), typically characterised by low-friction co-op experiences with a low-fi and/or kit-bashed aesthetic
The games are the popular, but I do not think that term to describe them is that well-used nor used with that much affection :)
It’s requires a good base group of friends that you will have fun with already, and the design for the game is “slop” / cheap / lazy in that regard. But not necessarily bad.
I've started writing all of my scripts in nushell (unless It's critical that they keep working long-term without maintenance). It's incredible, and improving fast.
Helix is a Vim/Kakoune-inspired modal editor, with a bunch of stuff built in by default. For example it has support for a huge amount of LSPs and intergrates them automatically.
It's command structure is also super similar to Vim's, but, basically, "flipped" around. So you wouldn't write "dw" to delete a word, but "wd". This means that you can see whatever you're selecting to be deleted highlighted before you actually execute the deletion. It has a bunch of saner commands also for stuff people usually want to do, like go to definition/usage, and honestly for people who aren't Vim-addicts such as myself, it's probably a good idea to check it out once, to see if it's a good fit for you.
I tried getting into nvim (handy for editing from the CLI or over ssh), but within weeks the plugin system started getting weird errors.
Then I tested Helix[0] when a friend suggested it and it Just Works. Along with LSP support that just picks up language servers automatically if you install them.
The target-action command style takes some getting used to after (n)vim's action-target style, but I actually prefer it now.
This feels like advice for the most basic level of empathy (support your friends) with the cherry on top of AI slop as the opener. Americans sure are something else.
It is funny that developers are always looking at the processes of car making to improve their own extremely broken ways and now it is car-makers that "should" be trying to be more like the agile software devs.
As a dev the last thing I want is a software-defined car. Look what we did to TVs.
