You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
My goal was to understand if this new tech (speckit) can be used beneficially. My hope was that it keeps the context brain rot under control but I guess I was mistaken to expect that. I tried to build a simple overleaf plugin (equation editor like in word). Since 5 days, I am trying to get the AI to build this simple module. I am kind of losing the rest of my mind, since the copilot with spec kit loses its marbles over time. I thought that the idea of the spec kit was to prevent this with the .md file?
I defined a skill set (I gave it two ways how to build the overleaf module: A. modify the source code, build the dockerfile and restart the docker container B. Hotfix files inside the docker container, compile the website inside and killall -2 node). In the beginning both worked fine. Then it started to compile the source code directly and destroyed the source code. I had to restore from my backups. I told it not to do it but it continue to do it anyway. Then it started to build and introduce a local server. I asked it again and again not to do that. It doesn't cared.
"We" agreed to have for every playwright test a separate sub-folder for the screenshots. In the beginning it played along. Then it decided to throw the screenshots all over the place.
I had 12 nicely defined playwright tests. The emphasis is on "had". After some time all the tests we doing the same, fusing into each other. Currently I am trying to convince it to accept my rule for the tests and rebuild them as they where. No chance.
I asked it now many many many times to not use env variables like NODE_PATH (because I can not auto approve them)
cd /data_1/dev_latex_editor/services/web/modules/latex-editor/demo && NODE_PATH=/data_1/dev_latex_editor/playwright-tests/node_modules node capture-minimized.js
It made me even a bash script. It used it 2 times or so and then went back to env variable.
And so on...
I would say that the experiment was a complete failure. An example of finest AI brain rot. Except it missed something important!?! In the end I asked Claude to help me write the messages for Copilot. Claude kind of "gave" up and told be to be firm with Copilot.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
-
My goal was to understand if this new tech (speckit) can be used beneficially. My hope was that it keeps the context brain rot under control but I guess I was mistaken to expect that. I tried to build a simple overleaf plugin (equation editor like in word). Since 5 days, I am trying to get the AI to build this simple module. I am kind of losing the rest of my mind, since the copilot with spec kit loses its marbles over time. I thought that the idea of the spec kit was to prevent this with the .md file?
I defined a skill set (I gave it two ways how to build the overleaf module: A. modify the source code, build the dockerfile and restart the docker container B. Hotfix files inside the docker container, compile the website inside and killall -2 node). In the beginning both worked fine. Then it started to compile the source code directly and destroyed the source code. I had to restore from my backups. I told it not to do it but it continue to do it anyway. Then it started to build and introduce a local server. I asked it again and again not to do that. It doesn't cared.
"We" agreed to have for every playwright test a separate sub-folder for the screenshots. In the beginning it played along. Then it decided to throw the screenshots all over the place.
I had 12 nicely defined playwright tests. The emphasis is on "had". After some time all the tests we doing the same, fusing into each other. Currently I am trying to convince it to accept my rule for the tests and rebuild them as they where. No chance.
I asked it now many many many times to not use env variables like NODE_PATH (because I can not auto approve them)
It made me even a bash script. It used it 2 times or so and then went back to env variable.
And so on...
I would say that the experiment was a complete failure. An example of finest AI brain rot. Except it missed something important!?! In the end I asked Claude to help me write the messages for Copilot. Claude kind of "gave" up and told be to be firm with Copilot.
:-(
Beta Was this translation helpful? Give feedback.
All reactions