RichKuo 🍇
Region: US
Wednesday 05 August 2026 02:51:50 GMT
2238
23
60
1
Music
Download
Comments
Fuck Google :
What stopped programmers from doing that before ai?
2026-08-05 12:25:30
3
WingDingKing :
hot take - most tests were completely fabricated before ai
2026-08-05 02:58:23
13
sg6 :
This is already the problem with unit tests, when you mock a bunch of objects the wtf are you really testing
2026-08-14 19:13:05
2
Ethan Fremen :
You force it to write tests first?
2026-08-10 00:47:15
1
The AI Syndicate :
You have an agent unaware of the coding session do an adversarial loop to verify the code. Never do the test in the same thread you coded in.....
2026-08-06 18:21:04
1
Crypto Bros Pls Leave Me Alone :
Do it twice
2026-08-05 21:32:12
2
Brian 🇺🇸 :
This is why I use openspec.dev and use a trusted model for /opsx:verify
2026-08-05 17:57:10
2
🎮Game_Artificer🎮 :
Why would you use the same AI for testing as implementation? If the AI could catch the error it made, it wouldn’t make the error.
2026-08-13 22:07:41
1
j :
what’s stopping it? Not much. Comes down to the quality of the RL environments they used on it in training and how resistant they are to reward hacking
2026-08-12 15:21:48
1
William :
Use different sesssions no shared context thisis much less likely
2026-08-08 17:20:23
1
Ace 🪐 :
I'm constantly telling Codex to stop writing useless and meaningless tests. Even as an AI scientist I know more about testing 😭 then it reports all tests passed! like testing if an pydantic attr is the specified type meant something
2026-08-06 02:06:38
1
mencuddling :
Always use another AI to check the work. For important PRs I have Claude Code Fable Mac do the work and benchmarks, then GPT 5.6 Sol and Kimi do an adversarial review. They always find stuff.
2026-08-08 19:13:26
1
DannLabs :
I have a claude command named, verify-task, what it does it gives a documentation if a specific task passed, with technical and human verification test
2026-08-07 00:22:08
1
patrickw :
I have the lm write the test before it builds the class/function etc , verify it fails, then build code, then test. I also have a diffrent llm make the test then the one that codes it using opencode. still can't get it to do good UIs and humanized workflows/good placement without multiple iterations though so open to feedback.
2026-08-05 03:22:02
1
chooven752 :
nothing. the same thing that stops you from doing the same.
2026-08-08 16:41:13
1
Francesco :
AI makes TDD finally worth it.
2026-08-06 06:23:41
1
ajaxbeats.backup1 :
use different models for questions and answers
2026-08-05 11:58:29
1
sg6 :
At this point unit tests are just a way to avoid typos, import error, and misclicks, not actually validating the code
2026-08-14 19:14:14
2
shalexxo :
The same thing that stops _me_ from doing the same exact thing lol
2026-08-05 03:22:27
2
✊🏼🏴 0dw01 🚫🍊👑 :
I have a code review skill that I run periodically to look for test gaps, tests that don’t test real functions of the app, duplicate tests, etc. It seems to help a quite a bit.
2026-08-05 17:34:35
1
To see more videos from user @rich_kuo1, please go to the Tikwm
homepage.
