Blog
Tutorials, build stories, and the things we learned along the way.
Top-K, Top-P, Temperature: Which to Change
Three sampling parameters, one clear decision rule. Know which knob to reach for first based on what your output actually needs.
Stop Watching AI Tutorials. Open a Playground Instead.
Watching AI tutorials is passive learning. You retain about 20% and never build real intuition. Here's why a playground beats 40 hours of AI videos.
AI Hallucinations: When Models Lie Confidently
Models hallucinate because they predict plausible tokens, not true facts. Here's why it happens, what makes it dangerous, and how to reduce it.
How to Get a Gemini API Key (Free)
Step by step: get a free Gemini API key from Google AI Studio, make your first call, and avoid the three mistakes most people make on day one.
Zero-Shot vs Few-Shot vs Chain of Thought
Three prompting techniques, three use cases, and the decision rule for picking between them. With real examples you can run in the playground.
System Instructions: The God Mode of LLMs
System instructions set the model's persona, format, and guardrails before the user types anything. Here's how to write them well.
We Told AI We Solved P=NP. It Believed Us.
We typed four obviously wrong claims into Gemini Flash Lite. It agreed with all four. Here's what AI sycophancy looks like in production.
How We Built TinkerLLM: 68 Exercises, 2 Wrong Turns
Full build story: 18 lessons, 68 exercises, React 19, Firebase, and Gemini. Including the two decisions we had to reverse before TinkerLLM actually worked.
Tokens Explained: How LLMs Read and Write
Tokens control API cost, context limits, and why your model gets cut off mid-sentence. Here's exactly how LLMs read and write.
What Temperature Actually Does in LLMs
Temperature controls randomness in AI output. Here's the math, the practical settings, and an experiment you can run yourself.