Fletcher Dunn

16.7K posts

Fletcher Dunn banner
Fletcher Dunn

Fletcher Dunn

@ZPostFacto

Engineer at Valve | Free 3D Math book: https://t.co/iPzvJWf0RC | GameNetworkingSockets: https://t.co/0U2Je5ZbYS | NOT doing support on discord!

Issaquah, WA Katılım Ekim 2014
1.8K Takip Edilen11.3K Takipçiler
Sabitlenmiş Tweet
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
3D Math Primer for Graphics and Game Development is now available in its entirety for free! gamemath.com Vectors & matrices, representing orientation in 3D, graphics & physics basics, and more. Beginner to intermediate level. Remastered to be web & mobile friendly!
English
52
366
1.3K
0
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
This is all hobby and opensource code BTW.
English
1
0
4
333
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
(Side note: IMO, these guys are experts in an extremely peculiar and narrow part of software, and their self-assessment is wildly inflated. They don't really even know what it is that they don't know. And it's telling how many of them are auteurs who don't work in teams.)
English
2
0
12
981
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
My impression of vibe coding after collecting a few more weeks of data: As coding/debugging sidekick it's *amazing*. A real killer app for this tech. It spotted a race bug by analyzing a failed test log. No-scope headshot from across the battlefield -- insane productivity increase.
English
8
2
87
6.8K
Danny
Danny@dcantah·
@ZPostFacto I think the biggest gain for me is it giving a hypothesis on various issues reported by users. What would previously take 20-30 minutes per issue in some cases, I can just have 5 term panes each investigating something. In most cases the hypothesis it gives is correct
English
1
0
1
15
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
@CLTetudo I ended up not taking the exact fix is proposed, but that's not the point, because typing in the characters of the change is 2% of the actual time. The analysis is 98% of the time. And then when I made my preferred fix, I had it analyze my fix. And it explained why it worked.
English
0
0
1
69
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
@CLTetudo That's a pretty detailed bug report, but it was able to explain the issue and confirm it's a real bug, and propose a reasonable fix. But it was great because I had a second opinion about the analysis.
English
1
0
1
73
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
@_trundler And this is why human engineers are probably not going anywhere, for a while at least because only an engineer really fully understands what those technical requirements are.
English
0
0
1
47
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
@_trundler You have to express your requirements in *some* language. At a certain level of technical detail in your requirements, the programming language offers the "best compression ratio" for communicating them. A well-tuned CLAUDE.md might help.
English
1
0
1
53
Fletcher Dunn
Fletcher Dunn@ZPostFacto·
@kiki___toki My experience is that it's best to be pretty "hands on" and express your criteria in as much detail as you can. And then don't let it go too far in one direction, without checking on the results.
English
1
0
1
335
DecafBagel
DecafBagel@kiki___toki·
Thanks for sharing your insight. I’m learning and trying to utilize it more in my workflow. Debugging, tooling and for one off things I don’t care that much or just need something working, it’s been really amazing. Hard to quantify the productivity boost. But for more complex, actual feature implementation, doesn’t matter if it’s gpt 6.9 ultra pro max, if I don’t review or watch it even for a bit, shit gets pile up real fast. People say prompt it better etc. skills, plugins, harness etc. I gave it a clear full paper, the og networking paper, and expect a e2e implementation. It failed and I had to get in there. web.cs.wpi.edu/~claypool/cour… Still learning how to utilize it better, love to see more insights from people I respect like yourself 🫰
English
1
0
0
364