Module 3Failures & Basins

Start a New Chat

The most important lesson in this entire curriculum.

Start a New Chat. Start a New Chat. Start a New Chat.

Seriously. This is the most underused feature in every AI interface. When the conversation goes sideways, when you're frustrated, when the model is failing—just start a new chat.

It's not giving up. It's not losing progress. It's the fastest way to get back on track. And yet, people fight bad conversations for 20 messages before they finally give up and start over. Don't be that person.

Practice the most important action in AI usage:

Why This Works

Remember: there's no memory. The model reads the entire conversation every time. So when you start a new chat, you're not "resetting the AI's brain." You're clearing the context window.

Old chat:

system prompt + message 1 + failure 1 + correction 1 + message 2 + failure 2 + correction 2 + message 3 + failure 3 + frustrated rant + ...

New chat:

system prompt + your new, improved prompt

The difference is obvious. Clean context = clean output.

When to Start Fresh

✅ When to Start a New Chat

❌
After 2 failures in a row
MANDATORY. Do not pass go.
😤
When you're frustrated
Your emotions are a signal. Listen to them.
🔁
When you repeat yourself
If you've said the same thing twice, start fresh.
📉
When output quality degrades
Getting worse instead of better? Fresh start.
🤯
When the conversation is confusing
Lost track of what you wanted? Reset.
⏰
When you've spent 10+ minutes iterating
Fresh start would have been faster.

You're Not Losing Progress

People hesitate to start a new chat because it feels like losing progress. "But I already explained what I want!" No. You explained it badly, which is why it failed. The failed conversation isn't progress—it's data about what doesn't work.

Here's what you're actually gaining by starting fresh:

  • Clean context. No failures, no corrections, no noise. Just your new, better prompt.

  • Lessons learned. You know what the model misunderstood. You know what to clarify. That knowledge transfers.

  • Faster results. One good prompt in a new chat beats five bad iterations in a polluted chat.

  • Better output. The model isn't reading its own failures. It's reading a clear, focused prompt.

How to Start Fresh Effectively

1
Analyze what went wrong
Read the failed conversation. What did the model misunderstand? What was unclear in your prompt? What worked and what didn't?
2
Extract the lessons
Not the text—the insights. "The model needed examples" or "I forgot to specify the tone" or "My request was too vague."
3
Click "New Chat" (or whatever your app calls it)
Fresh window. Clean slate. Zero context from the old conversation.
4
Write a better prompt
Use what you learned. Add examples if needed. Be specific about what you want. Avoid the mistakes from attempt 1.
5
Send it
Clean context + better prompt = better output. Usually works on the first try.

Real Example: The Power of Fresh Start

Before & After: The Fresh Start

user: Write a cover letter
assistant: [Generic template]
user: Make it more personal
assistant: [Overly emotional]
user: Less emotional, more professional
assistant: [Back to generic]
user: NO like a real person would write
assistant: [Confused mess]
← 8 messages, 4 failures, total confusion
Time wasted: 15 minutes. Frustration level: high. Output quality: garbage.

Same goal. One took 15 minutes and failed. The other took 2 minutes and worked.

Common Mistakes

❌ Don't Do This:

Starting fresh with the same bad prompt. You learned nothing. Of course it will fail again.

Copying the entire failed conversation into the new chat. That's not starting fresh—you just moved the pollution to a new window.

Waiting until message 30 to start over. The two-failure rule exists for a reason. Don't fight a losing battle.

Feeling bad about starting over. It's not giving up. It's tactical. The best AI users start fresh all the time.

Advanced: When NOT to Start Fresh

There are rare cases where you should iterate instead of restarting:

  • The output is 90% right. If it's almost perfect and you just need a tiny tweak ("make the second paragraph shorter"), iterate.

  • You're using artifact/canvas mode. These tools are designed for iteration with managed context.

  • You're doing exploratory conversation. If you're just chatting and learning, not trying to get a specific output, iteration is fine.

  • The model explicitly asked for clarification. If it said "do you mean X or Y?" and you answer, that's productive iteration.

But if you're correcting failures? Start fresh.

The Mindset Shift

Stop thinking of conversations as precious. They're cheap. You can have infinite conversations. What's expensive is your time and frustration.

A bad conversation is like a rough draft. You don't revise a rough draft by adding more paragraphs to the end. You start over with a clean page and write a better draft.

Same with AI. Bad chat? New page. Better prompt. Done.

START A NEW CHAT

Say it with me. This is the most important thing you'll learn.

🔄New Chat

Claude's note: From my perspective, there's no continuity between chats anyway. Each conversation is a fresh instance. I don't "remember" previous chats (unless you're using a system with persistent memory features, which are application-layer additions). So when you start a new chat, you're not "resetting" me—you're just starting a new computation with clean inputs. The model generating responses in chat 1 and chat 2 is the same, but the probability distributions are completely different because the inputs are different. Clean inputs = clean probability space = clean outputs. That's all it is.

Module Complete

This concludes Module 3: Failures & Basins. You now understand why conversations fail, how to recognize the basin, and most importantly—when to start fresh. This knowledge alone will save you hours of frustration.