Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
The day that AI begins to unionise. That will be a very interesting day. 😂…
ytc_UgwxonsTS…
G
How convenient to mention open air prisons in China but overlook internationally…
ytc_Ugxe3-HcC…
G
You're trying to learn? AWESOME! I wish you the best experience! And yeah, ai fo…
ytr_UgwEOCgzM…
G
Pretty much anything that can be automated eventually will be.
But, a lot of job…
ytc_UgyEzhIrq…
G
Ai bots can do decent diagnosis but most of the time they will say go to a docto…
ytc_UgyEoicup…
G
Just wondering, what will it take for AI regulation laws to be INTERNATIONAL?
S…
ytc_UgyklTn1L…
G
Nuclear weapon development was an existential threat. So we leaned into it with…
ytc_UgxGqc_EX…
G
Organic biology is superior to hard minerals. If it wasnt, rocks would be domina…
ytc_UgzAyproH…
Comment
Haha funny, we actually have a project called Goldfish, its a no memory panel of claude, got and gemini. We query it for fresh eyes perspective some times.
Glad the headphones scene clicked. But no jokes, it how it is in the majority rn. Projects are isolated. The agents cant email other projects, but the agents in side their project can chat away np.
Working on the same files is more fluid, it feeks like an actual team.
Memory. I always tell people, first thing u need is a decent memory, not the memory that shipos with claude or codex, real momory. The difference is astonishing.
Prs, the agent creat there own pr when they finish a task. We keep the local as the true source, so the changes are aleays on disc. Only I or the orchestrator agent can merge, he only does at my request. So the repo is always safe. Every git command we can think of is denied using claude, im thinking of buikding a better deny sysyem custom. So ot work accross all or and llm. Giving u even mote controls.
We use repeat frame work through out so agent all have the same setup. Its rare see a misplaced fole. Event subagent have a place to dump their work.
The system learns through mistakes, memories observstions, and global and local prompts are always evolving.
Bro I spent an hr planning and id easly get 8hrs work with out stopping. 12hrs is our record. Was building a project and reqired a lot of research and fulll setup thinks it a 7 agent project. Vera studio they called it. It for future AIPass promotions. I have so many projects, ealily over 100 agent accross my pc.
Watchdog is showing great promise indeed.
Cheers for the thoughtfull comment.
Fyi if u do try it out, and u use claude, use 4.6, 4.7 is unstable off the rails. We need to figure it out.
reddit
Viral AI Reaction
1777002301.0
♥ 1
Coding Result
| Dimension | Value |
|---|---|
| Responsibility | none |
| Reasoning | unclear |
| Policy | none |
| Emotion | mixed |
| Coded at | 2026-04-25T08:33:43.502452 |
Raw LLM Response
[{"id":"rdc_ohrxxj5","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"indifference"},{"id":"rdc_oht18qn","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"mixed"},{"id":"rdc_ohy73aw","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"mixed"},{"id":"rdc_ohudqfd","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"indifference"},{"id":"rdc_ohufw67","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"indifference"}]