mastodon.gamedev.place/@Steverock

32.2K posts

mastodon.gamedev.place/@Steverock banner
mastodon.gamedev.place/@Steverock

mastodon.gamedev.place/@Steverock

@steverockan

AI apps @ google; past: eng lead @Subnautica, Game Builder https://t.co/upyvgJuTT0, google poly gfx (RIP), https://t.co/gxcEO6Q6hl; he/him

NYC Katılım Şubat 2009
496 Takip Edilen799 Takipçiler
Fernando
Fernando@Franc0Fernand0·
I don't understand how you can get good at building with AI without learning to code first. If you've never built things by hand, you won't know: •What "good" actually looks like •When the AI is wrong or hallucinating •What to ask for next You're not directing the AI. You're just hoping it works.
English
117
17
247
16.6K
Denislav Gavrilov
Denislav Gavrilov@kuberdenis·
ai is not coming for your job ai stripped you from debugging a ci failure for 3 days, or 4 hours of reading/trying out a library's exported functions, exploring performance steps, or wiring a strange protocol ai just made it so you never ever enjoy intellectual work ever again
English
81
26
999
99.5K
Sebastian Aaltonen @ SIGGRAPH
Some might think that Kimi K3 open source model catching up with OpenAI and Anthropic is a clear sign of the AI bubble breaking. I disagree. Kimi K3 ran out of AI server capacity immediately. They need to buy more hardware. Kimi K3 is also open source. Companies are downloading it and setting $100k AI servers inside their intranet. AI hardware companies will be fine. They are doing even more money. OpenAI and Anthropic are not yet publicly traded. Their stocks can't collapse yet. I wouldn't personally buy them. Too much risk. Not a proven business model.
English
15
2
125
12.1K
mastodon.gamedev.place/@Steverock
@Grummz @Jonathan_Blow I will bet you $10,000 that that day will not come in 5 years. 100% serious, DM me. We can do it publicly, we of course need to spell out the details, but I’m willing to make this bet.
English
0
0
0
44
Grummz
Grummz@Grummz·
@Jonathan_Blow That day is on an exponential improvement curve, not linear.
English
14
1
31
3.6K
Grummz
Grummz@Grummz·
Next year or so, only game designers will matter. It's clear that the cost of software coding is zero. Any feature, of any complexity is on tap right now, and will get cheaper. After that all that will be left is design, taste and distribution. Very human tasks still.
English
144
50
1.1K
55.5K
mastodon.gamedev.place/@Steverock
Using AI image gen as part of the process has been a huge breakthrough in my quest to generate DOOM levels
English
0
0
0
69
mastodon.gamedev.place/@Steverock
i vibe coded a script to render WADs into these heat maps, then asked GPT image to create variants. now to see how well claude can convert these height maps to playable WADs..
English
0
0
0
19
mastodon.gamedev.place/@Steverock
taking a different approach to AI-powered doom level gen...the automaps are probably not the best intermediate rep (image gen -> intermediate -> convert to WAD), so trying a heat/height map instead:
mastodon.gamedev.place/@Steverock tweet media
English
1
0
0
58
mastodon.gamedev.place/@Steverock
@sebify yep. i think AI has a tendency to create very hacky code. it often feels pretty brute-force, like it's just greedily searching the solution space.
English
0
0
0
36
Sebastiano Mandalà
In my 25+ years of coding I realized a long time ago that engineers have an incredible hard time to visualize how to correctly create architectures. Most of the times are over engineered and inefficient. There will be very few examples of good architectures that AI can learn from
English
4
0
7
557
Sean Sanker Jr
Sean Sanker Jr@SeanSankerJr·
hot take - software quality will actually go up over time AI makes it easier than ever to improve code, run adversarial review, add tests, and explore previously infeasible migrations sorry not sorry
English
58
5
189
11.3K
Ryan Florence
Ryan Florence@ryanflorence·
Talked to my friend who owns a small business He replaced their CRM with one he built with Claude Code a few months ago He is not a programmer, has never seen the code, doesn't know what stack it used It runs on a mac mini in their office "It's way better than what we had"
English
40
5
335
38K
Ryan Florence
Ryan Florence@ryanflorence·
@DustinTownsend mac mini in the office doesn't have to worry much about security aside from being stolen from the office.
English
3
0
3
1.1K
Tanuj
Tanuj@tanujDE3180·
Backend Interview Question: A cron job should run once at 2AM Today it ran twice. No manual trigger. No duplicate servers. No retries. What happened?
English
60
6
166
39K
mastodon.gamedev.place/@Steverock
@MarkJSzymanski Yeah, vibe coding static websites seems like a pretty good idea. They’re relatively simple, I wouldn’t worry about maintainability much. If we’re talking about a blog system, I think using a system is probably still good, but ideally they’d have a SSR solution
English
0
0
1
18
Mark Szymanski
Mark Szymanski@MarkJSzymanski·
Terrific question. From 2016 to early-2025, one could have made an argument that building marketing websites with some sort of website builder (WordPress, Wix, Squarespace, etc) was an overall better approach. With the advent of agentic coding, I am no longer feeling like that's the case. Static sites are not entirely perfect when you're trying to build for clients, but the cost-benefit analysis has shifted significantly for my business.
English
1
0
0
139
Mark Szymanski
Mark Szymanski@MarkJSzymanski·
It's amazing how much bullshit just evaporated when I started building marketing websites statically. Whole categories of problems I used to fight with. Gone. - No database to get hacked - No plugins to update or that break each other - No security plugin - No caching plugin - No server to go down - No backup plugin, it's just on GitHub - Way less money every month Just so much less shit to worry about.
English
41
4
118
16.3K
Ryan Brewer
Ryan Brewer@ryanbrewer·
@kyledcollins Most of that is solved. I miss the flow state of not having to outsource thinking and managing 10 threads
English
3
0
13
1.2K
Kenton Varda
Kenton Varda@KentonVarda·
Yeah exactly. I've reviewed a lot of human code in my career. Every company I have worked at over the last 20 years has been an "every change needs to be reviewed before merge" type of shop. And in the process I've caught an enormous number of bugs and logic errors. Almost everything I review has problems. Reviewing AI code honestly doesn't feel much different to me than human code ever did. Most of the problems I catch feel just like problems I've seen in human code. I feel like the people complaining about AI code quality must never have been in a role where they are regularly reviewing their colleagues' code?
tobi lutke@tobi

A huge amount of the Anti-AI code sentiment massively overestimates the quality of human code outside of a very small set of open source and high quality company codebases. Human Slop is everywhere and can trivially be improved on by any opus level model.

English
106
96
2.1K
188.4K