Strangely Useful desk

AI Tools

Plain-English AI tool guides covering what each tool does, where it helps, where it fails, and how to use it responsibly.

A document is redacted before entering an AI chat.

AI Toolsstory6 min

Check AI Chat Training and Retention Before You Paste

Choose between a training opt-out, temporary chat, history deletion, and memory controls without assuming one switch erases everything.

A redacted document copy is uploaded while the master remains stored.

AI Toolslist7 min

Before You Upload a File to ChatGPT, Choose Where It Will Live

A file can remain in Library after its chat is deleted. Choose the right upload location, remove unnecessary data, set training controls first, and clean up every stored copy.

An AI task runs locally and in the cloud with different boundaries.

AI Toolsstory2 min

Local AI Keeps More Data on Your Device—but Local Does Not Mean Risk-Free

Local inference can avoid sending prompts to a host. The app, plugins and device security still define privacy.

A hidden webpage instruction is blocked from agent tools.

AI Toolsstory2 min

Prompt Injection Is the Phishing Problem Inside an AI Agent

A malicious page can contain instructions aimed at the AI reading it. Limit permissions and confirm actions.

AI claims are checked one by one against primary sources.

AI Toolslist2 min

AI Citations Are Leads, Not Proof

A real link can still fail to support the sentence beside it. Open, match and contextualize every consequential claim.

Only selected passages stay in focus within a long AI context.

AI Toolsstory2 min

A Bigger Context Window Does Not Mean the AI Read Everything Well

Capacity measures what can fit, not whether every detail receives equal attention.

Participants see notice before AI meeting capture begins.

AI Toolslist2 min

AI Meeting Notes Need Consent Before They Need a Summary

A meeting bot can capture voices and decisions. Tell participants and provide a non-recorded path.

AI-generated code passes through review checkpoints.

AI Toolslist2 min

Treat AI-Written Code Like a Pull Request From a Stranger

Generated code can contain insecure defaults and invented packages. Review the diff before running it.

A media file is checked against its source, provenance, and a bounded detector signal

AI Toolsstory7 min

A Deepfake Detector Score Is a Lead, Not a Verdict

Preserve the original, verify the exact claim and source, check provenance, and use a detector only when its medium and evaluation conditions fit the file.

An image carries a signed timeline of creation and edits.

AI Toolsstory2 min

Content Credentials Explain a File's History—When the History Is Present

C2PA credentials attach signed creation and editing history. They do not judge whether a scene is true.

An AI summary is checked against the original report.

AI Toolsstory2 min

AI Summaries Save Time Only If You Keep the Original Nearby

A summary can omit a condition or flatten disagreement. Check consequential points in the source.

An AI agent receives a small set of access keys.

AI Toolslist6 min

Limit an AI Agent's Permissions Before It Acts

Map the job, separate read from write, require approval for consequential actions, and know which control actually removes access.

An AI archive is checked before account access is closed

AI Toolsstory8 min

Export and Delete Your ChatGPT or Gemini Data in the Right Order

Exporting is only a snapshot. Verify the archive, remove separate copies, cancel the right subscription, and use the correct personal-account deletion path.