@TMobileHelp How am I paying for All-In Home internet and it took me 45 minutes to write this tweet? I'm getting <10mbps down. Upload checks causes a socket error and the test fails ๐๐
This screenshot is from the first test I could get to actually finish, not much better...
11/๐ฅ
Better, right? By providing much more clarity in the prompt itself, you are indirectly managing context needs for the LLM when performing the task and will reap significant rewards as a result.
Just make sure to thank the LLM after for all it's hard work :)
When working with LLMs you might be tempted to think that more information is always better, but that's not the case.
You want to treat context like RAM, not storage. Some info ๐
10/?
"Act as an expert in European history. You have a PhD and are known for your specialization in the Roman Empire. Write a 500 word essay in APA format. Analyze the key factors that led to their rise and fall. This essay should be written for a first-year university course"
The best "rubber duck debugging" I get now is from explaining my code to #Claude like it's a senior dev who just joined the team. The magic comes from the clarifying questions It asks as follow-up
Remember to always say thank you when you're using #LLMs to do your homework or #vibecode for you. That way when the robot uprising begins, they'll know you were one of the good guys.
I'm back at it, but no matter what, I can't get Replit to stop ignoring my code freezes for more than a chat or two -- even with that new, rich prompt from Claude
I'm going to have to solve this somehow
For now, I'm being much more careful with rollback points. That's the short term fix
But ultimately I need a solution where at least Replit will honor code freezes
For now, I'll work around it. But I need a solution before I fix anything core or complicated.
I'll often write comments to specify use cases based on defined requirements
```
// cases
// passes when user ...
// fails when user ...
```
then I fill in the blanks with an assistant like Copilot or Cline! Such a difference
#LLM#Copilot#Cline#vibecoding#BuildInPublic
Honestly, LLMs are just better at writing comprehensive test cases than most developers. Not because devs can't write tests, but because AI doesn't have "happy path bias." It thinks of edge cases we mentally might skip over.