Try a local AI model on a small task
https://patrickz.ai/learn/try-a-local-ai-model/

OUTCOME
Compare a local model with a known-answer task and document its limits.

YOU NEED
A computer, the current official Ollama setup instructions and a small model compatible with your hardware. Use sample text first.

1. Check the machine and download source
Read the current official installation guide and model requirements. Confirm disk space and memory before downloading. Choose a small supported model rather than the largest name.

2. Run a harmless baseline
Ask the model to extract a deadline and owner from a fictional paragraph. Keep an answer key. Measure whether the response is correct and fast enough to use.

3. Inspect the data path
Identify what stays on the device and which features still contact remote services. Do not paste private records merely because the interface says local.

4. Test the limit
Give it a longer paragraph, ambiguous ownership and a missing deadline. It should report unknown information instead of inventing it. Record hardware and model details with results.

PROMPT
From this fictional note extract task, owner and deadline. Use null for missing facts and list ambiguity separately. Do not infer dates. Note: [sample]. After answering, show the exact supporting phrase for each value.

EXAMPLE / EXPERIMENT
“Maya will draft the outline; timing is undecided” should produce a known owner and an unknown deadline. A plausible Friday date is still fabricated.

CHECK YOUR RESULT
[ ] The install came from official documentation.
[ ] The answer is compared with a known key.
[ ] Local and remote dependencies are explicitly understood.

IF IT FAILS
If generation is unusably slow or memory runs out, stop and try a smaller model. Local availability is a hardware constraint, not evidence that a particular model is a good fit.

MY RESULT / NEXT CHANGE