@gmiller IA just accelerated things, IA is not the problem. Think about it, the confusion between evaluation and learning is the core issue. There is no "cheating" and no gain in offloading work to IA if evaluation is separated from learning. I am working on an app that fixe that.
Creating and managing people isn't the same thing at all. And I am tired of people thinking devs are just code generating machines, I create, I make, from day one of me learning how to code in 2015, coding is just a tool I use, I am not the tool. (2/2)
"Delegation is becoming available to those who never intended to manage (rather than do the work)." Well said. I sometimes hear manager thinking IA will replace dev, but what will a manager with no dev be? A maker with dozens of IA agents? (1/2)
The more I think about this, the more I feel I am developing cognitive dissonance.
Delegation is becoming available to those who never intended to manage (rather than do the work). Coders, designers, artists. Each hates and loves this tech for different reasons.
Their lament is one of atrophy, of losing connection to the work.
Yet delegation is also a mode of power. It's seizing the means of production and, for lack of a less loaded term, exploiting GPUs to do the work instead. Call it quantum leaps in increased productivity, slop cannons, or anything in between... but it is ultimately a move of empowerment.
So we're gaining power by not having to do the thing we're trained to do, and in so (not) doing, we lose the skills we still value so much more than being able to conjure work from thin air.
Maybe wizardry IS a good analogy here. We're summoning real-world products into being, works of fiction, images, videos... ethereal agents, familiars, are doing our work for us.
Does the wizard who conjures fire with the flick of a wand regret not having to buy matches?
I feel like I’m frying my dopamine system with all this Claude/Codex work.
Constantly monitoring and instructing, anticipating the work it’s going to produce.
Then killing time waiting, consuming way more algo feed content than ever before.
How do y’all tackle this?
@theo_the_dev Because if it requires or could benefit from changes, then you have your answer already, if you really meant a perfect first (lazy) prompt result, than how about applying that powerful IA to cure all diseases and make spaceship instead?
@theo_the_dev I need a more precise statement to respond, does that theoretical game you mention benefits from a game dev review/extra prompt s, or is it perfect from the first prompt and no changes make sense ? (1/2)
I also express my opinion as a question to IA and I get better results. Make me think that I read somewhere that "strong confidence is often confused with strong knowledge, when the opposite is often more true", yeah, experts doubt.
it's impressive how much more useful models become if you don't pretend to know the best way to do things.
e.g.
"refactor this huge class to use dependency injection"
is MUCH worse than:
"would it make sense to refactor this class to use dependency injection? what are the tradeoffs?"
true about life too btw
@kaiju_kam The latest gen of models as of 3 weeks ago are extremely good at understanding binary. This is making it super easy to create mod files, decomp games, and recomp games. At the rate of improvement, basically every game is going to be effectively open source in the future.
Some companies dislike their devs and artists (seen as a cost instead of an investment). And have not understood what can be reasonably be made by IA, and what reduces quality.
My problem with AI right now is that it isn't being used for what it should be.
In games AI could automatically generate forests, mountains and clouds etc. or do busy work that would be labor intensive/not that important (you can always give it a human touch/handcraft everything after), it could handle weather, NPCs actually having dialogue/conversations that fit their character, or the AI can just do busy work, 3D modeling is labor intensive, maybe you get asked to make a barrel, and normally it would take you an hour to get all the details right, but the AI could do the dirty/work that doesn't matter (like making the barrel shape etc) and 1 hour goes to 5 minutes, you spend 30 minutes personalizing and putting in even extra details, in the end it took you 35 minutes and you saved 25 minutes, which you can use to pour your heart into the game where it matters. The possibilities are really endless.
And that's the problem, too many companies only care about John Shareholder and the bottom line and fail to see the bigger picture, instead of using AI to help development, they try to use to replace Developers, instead of hiring a real artist, they'll let AI churn out absolute slop which is inherently soulless/unappealing, AI can do "busy work" and help artists find references (real ones) faster, thus speeding up development.
But of course humanity can't have nice things, so they are pumping trillions of dollars into AI and data centers which are a massive drain of money/resources and DO NOT generate a sustainable profit/break even. (butter not worth the spread) It's insane to me we are setting up the industry to fail (Dot Com Bubble 2.0) which will collapse the stock market/global economy.
AI is meant to be a useful tool, not a replacement for human jobs.
Recent AI developments are speeding up game decompilation.
Several projects are already making major progress, with some reaching 100%, most projects are using AI agents to handle parts of the reverse-engineering process
A fully decompiled game can then be rebuilt and lead to native PC ports, better mods and other fan projects.
@rmoff I know the feeling, when I make the effort to write things humanly, short, concise and digestible and all I get is a huge lump of text vaguely answering (or not answering at all) which clearly show how little effort has been made to respond.
@rmoff Obviously, it depends on what the conversation is about, your LLM getting another LLM text from a coworker with different access and context to help fix an issue is not first meant for human read. If the text is meant for human raw LLM output has the issues you mentioned (2/2).
@rmoff I have experienced that aswell. My opinion is that your LLM can talk with their LLM, but if they want it read by you, the human, they have to write it short and human digestible. (1/2)
IA just accelerated things, IA is not the problem. Think about it, the confusion between evaluation and learning is the core issue. There is no "cheating" and no gain in offloading work to IA if evaluation is separated from learning. I am working on an app that fixe that.
Conservatives need to understand this.
AI is already ruining higher education. It's already been far more disruptive, toxic, and demotivating than wokeness & DEI ever were.
Academics can't assign homework, take-home problem sets, or term papers -- too many students will use AI to cheat.
We can't give assignments or run assessments in online courses -- too many students will use AI to cheat.
We can't be sure what percent of any honors thesis, masters thesis, or dissertation is really the student's own work. Likewise for grant applications, grad school applications, and journal papers.
Even worse, after AI-enabled 'smart glasses' with cameras and microphones become common and inconspicuous, we won't even be able to run in-class quizzes, tests, or discussions without worrying that students are secretly using their AIs to feed them answers and prompt their discussion questions.
Wokeness was a serious threat to the intellectual integrity and viewpoint diversity of universities.
But AI is a serious threat to the entire academic enterprise.
If you're a conservative and you care at all about classic education and the Western canon, you need to wake up to the real threat here -- students offloading all reading, writing, learning, and skill development to AIs.
@gmiller IA just accelerated things, IA is not the problem. Think about it, the confusion between evaluation and learning is the core issue. There is no "cheating" and no gain in offloading work to IA if evaluation is separated from learning. I am working on an app that fixe that.
@IT_unhinged How is that greed ? They use it for work and even pay for it from their own pocket. Getting it approved is most likely too much paperwork and a long process, to just end up with a no response.
@edykstein Intéressant cette analyse. Par contre, si on prend la position que un interview c'est dans les deux sens, je doute fortement que le/la RH apprécie si le candidat demande "Me recruterez-vous quand même si je ne venais pas bosser ?"
@JFPuget In my opinion, sharing IA output to an human, without validating it first, is shifting the work on someone else, and in most cases, unrespectful.
@Fnao32978420 Well, IA just do some parts of it, it all depends on how much human there is in the loop. Coding can be done by IA and it makes no difference for a user. The art (visual/sound mostly) is where it can feel soulless. Even so, there is a difference between IA arts.