Assignment Review

Q1: Why do tokens matter?

Tokens are the units processed by an LLM. Every prompt, log file, YAML file, or response consumes tokens. More tokens increase cost, latency, and may exceed the model's context window.

DevOps Example

100 log lines = fewer tokens
10000 log lines = more tokens

More tokens = more Bedrock cost.

---

Q2: Context Window is similar to?

RAM

Server RAM
      ≈
LLM Context Window

If RAM is full → problems.

If Context Window is full → older information gets dropped or request fails.

---

Q3: Prompt vs System Prompt

Prompt is casual user input but system prompt decides behavior

Type	Purpose
Prompt	User request
System Prompt	Defines AI behavior and rules

Example:

System:
You are a Kubernetes Expert.

User:
Why is my pod restarting?

---

Q4: Temperature

Your answer:

Low

✅ Correct.

Reason:

Troubleshooting
RCA
Automation
Runbooks

must be consistent.

We don't want AI being creative when diagnosing production issues.

---

Q5: Hallucinations

Hallucinations are dangerous because AI may generate incorrect troubleshooting steps, wrong root causes, or non-existent commands. Following such advice in production could worsen incidents.

Example:

AI:
Run kubectl delete namespace production

if hallucinated → disaster.

---

Mini Project Review

You are a Senior Kubernetes SRE with 8 years of experience.

Analyze the Kubernetes pod failure using the logs below.

Pod Name: <pod-name>
Namespace: dev

Logs:
<logs>

Provide:

1. Root Cause Analysis
2. Severity (Low/Medium/High/Critical)
3. Probable Cause
4. Impact
5. Recommended Fix
6. kubectl Commands to Validate
7. Preventive Measures

Return the response in markdown format.

This is much closer to what enterprises use.

---

Module 2 Status

✅ Tokens

✅ Context Window

✅ Prompts

✅ System Prompts

✅ Temperature

✅ Inference

✅ Hallucinations