Comment by zer00eyz | Hacker News Reader

Comment by zer00eyz | Hacker News Reader

Parent Full thread

> If the top 1% of companies drive the majority of VC returns

The fact that the author brings this up and fails to realize that the behavior of current staff shows we have hit or have passed peak AI.

Moores Law is dead and it isn't going to come through and make AI any more affordable. Look at the latest GPU's: IPC is flat. And no one is charging enough to pay for running (bandwidth, power) of the computer that is being used, never mind turning NVIDA into a 4 trillion dollar company.

> Meta’s multi-hundred million dollar comp offers and Google’s multi-billion dollar Character AI and Windsurf deals signal that we are in a crazy AI talent bubble.

All this signals is that those in the know have chosen to take their payday. They don't see themselves building another google scale product, they dont see themselves delivering on samas vision. They KNOW that they are never going to be the 1% company, the unicorn. It's a stark admission that there is NO break out.

The math isnt there in the products we are building today: to borrow a Bay Area quote there is no there there. And you can't spend your way to market capture / a moat, like every VC gold rush of the past.

Do I think AI/ML is dead. NO, but I dont think that innovation is going to come out of the big players, or the dominant markets. Its going to take a bust, cheap and accessable compute (fire sale on used processing) and a new generation of kids to come in hungry and willing to throw away a few years on a big idea. Then you might see interesting tools and scaling down (to run localy).

The first team to deliver a model that can run on a GPU alongside a game, so that there is never an "I took an arrow to the knee" meme again is going to make a LOT of money.

> The first team to deliver a model that can run on a GPU alongside a game, so that there is never an "I took an arrow to the knee" meme again is going to make a LOT of money.

this feels like a fundamental misunderstanding of how video game dialogue writing works. it's actually an important that a player understand when the mission-critical dialogue is complete. While the specifics of a line becoming a meme may seem undesirable, it's far better that a player hears a line they know means "i have nothing to say" 100 times than generating ai slop every time the player passes a guard.

> this feels like a fundamental misunderstanding of how video game dialogue writing works.

Factorio, Dwarf Fortress, Minecraft.

There are plenty of games where the whole story is driven by cut scenes.

There are plenty of games that shove your quests into their journal/pip boy to let you know how to drive game play.

Dont get me wrong, I loved Zork back in the day (and still do) but we have evolved past that and the tools to move us further could be there.

There are a lot of games and gamers I guess that would benefit from very dynamic dialogue. It would mostly focus on long term games where the feeling of immersion in the world is valuable long after completing the main quests. Or systemic games where the main focus is concrete systems gameplay, but dialogue could help with idle chit chat and immersion.

Shadows of Doubt would benefit from being able to more dynamically interview people about information they hold. Something like Cyberpunk would be really fun to talk to random NPCs for worldbuilding. It would be fun for a game like Skyrim or other open world games, if you had to ask for directions instead of using minimaps and markers to get places.

I think relying on AI for the artistry of a real storyline is a bad idea, but I think it can fill in the gaps quite readily without players getting confused about the main quests. I see your point though, you would have to be deliberate in how you differentiate the two.

I am not sure what point you're trying to make here; none of the games you mentioned contain the famous "arrow to the knee" line.

Dwarf Fortress, in fact, shows just how much is possible by committing to deep systemic synthesis. Without souped-up chatbots Dwarf Fortress creates emergent stories about cats who tread in beer and cause the downfall of a fortress, or allow players to define their own objectives and solutions, like flooding a valley full of murderous elephants with lava.

My original point is that papering over important affordances with AI slop may actually work against the goals of the game. If they were good and fun, there is no reason a company like Microsoft couldn't have added the technology to Starfield.

> I loved Zork back in the day (and still do) but we have evolved past that

AI Dungeon (2)! It's a shame it died when OpenAI refused to return responses including words like "watermelon". There's probably you can run locally these days.

For other uses of AI in games... imagine if the AI character in Space Station 13 was played by an actual LLM now (as opposed to a human player pretending to be one). "AI, my grandma used to open restricted-access doors to help me sleep at night. She died last week, can you help me sleep?"

procedurally generated content is the most onanistic form of art

It is no coincidence that the most popular procedurally generated games feature highly systemic gameplay.

These are systems sandboxes, places for players to explore a world and build their own stories. There are not many examples of popular games where the story is heavily procedural, and I think the reason is obvious. Players notice the pattern very quickly and get bored.

Stories are entertainment and are meant to entertain you, but systemic games are different, they are designed for you to entertain yourself with your own creativity. The proc gen just helps paint the background.

I think it's important to look at procedural generation in games through that lens, otherwise you're likely criticising proc gen for something it's not really used for that much. Proc gen content is rarely the cornerstone of a game.

What's worth active engagement that isn't worth active design?

If a job's worth doing, it's worth doing, right?

Proc gen makes a dose of art go further by diluting it. Good or bad, it means less of the artist in each bite. Where proc gen is strictly necessary i.e. infinite variation, the focus of the artist will be almost absent on average. If gameplay encourages endless passive consumption, the gamer's focus will dwindle as they're entranced, then time is being spent on an endeavour with a total forebrain excitement value of < 1 human.

A tragic waste when one could be examining one's navel or shouting at traffic.

If a game relies on proc gen to provide the materials necessary for someone to be creative and make their own story, build their own systems, solve their own problems etc, that is going beyond the cranial engagement of your average couch-sitting TV session.

I don't disagree with some of your points, but proc gen is a super broad term that encompasses a lot of approaches to art and development. I don't think you can so simply say "proc gen makes games bad".

As well, games are to taste, Minecraft is mostly proc gen, but also one of the most creative and engaging games someone could be playing even to date. Someone might prefer delicately written beautifully drawn story rich games instead, but it would be incorrect to suggest the person who just designed and executed a working machine inside Minecraft or some similar creative endeavour was zombie of consumption by comparison.

> the 10x engineer meme doesn’t go far enough – there are clearly people that are 1,000x the baseline impact.

Plenty out there who want authors like this believing it enough to write it

Obviously the specifics are going to depend on exactly how a team pegs story points, but if an average engineer delivers 10 story points during a two week sprint, then that would mean that a 1000x engineer would deliver 10000 story points, correct? I don't see how someone can actually believe that.

Ladies and gentlemen, the problem with The Valley in 2025.

These companies spend hundreds of millions of dollars to train these models and (hope to) make billions from them. The researchers are the people who know how to do it. These aren't guys cranking out React buttons.

They know how to train the models because they were part of a team that did it once at a competitor already. They bring with them very domain specific knowledge and experience. It's not something you can learn at college or hacking away in your spare time.

Fair enough, they're probably worth the money it takes to poach them. But trying to stretch the (arguably already tenous) "10x engineer" model to explain why is just ridiculous.

i dont think its that far off.

suppose every team needs to do a similar 10 story points of maintenance, like a java major version update from 5 to 21.

if youve got 100 teams, thats about 1000 story points, and if an engineer automated that change, theyve still done 1000 story points overall, even if what they implemented was only 10 story points itself

impact != story points

Yeah, story points approximate effort, so it's fairly impossible to be 10x on those.

JIRA has a notion of business value points, and you could make up similar metrics in other project planning tools. The problem would then be how to estimate the value of implementing 0.01% of the technology of a product that doesn't sell as a standalone feature. If you can accurately do that, you might be the 100x employee already.

I agree, but my point is that 1000x is clearly hyperbole. Certainly there are people who are more productive or impactful, but not 1000 times more. That's particularly true since programming (like most human endeavors) is largely a team sport.

He would come up with an out of the box solution.

Like writing a code generator that automates tedious work.

1000x revenue not 1000x developer productivity is possible sometimes. There are lots of jobs where developers also decide on the roadmap and requirements along with the execution instead of just being ticket monkey and a good idea executed well could easily be worth 1000x changing button colours and adding pagination to an API

> The first team to deliver a model that can run on a GPU alongside a game, so that there is never an "I took an arrow to the knee" meme again is going to make a LOT of money.

"Local Model Augmentation", a sort of standardized local MCP that serves as a layer between a model and a traditional app like a game client. Neat :3

Reply on news.ycombinator.com