GPT-6 Astra vs Sol vs Luna: Which Tier Fits the Work and Budget?
Choose between GPT-6 Astra, Sol, and Luna using current API prices, official workload guidance, and a matched four-task Codex test that found no quality winner.
11 primary guides
AI Fundamentals
Core concepts and practical foundations for using modern AI systems responsibly.
Start with a job
Guided collections in this topic
Main subject
AI Fundamentals guides
These guides have AI Fundamentals as their primary topic.
How to Audit Agent Permissions Using OpenAI's Model Misalignment Reports
Audit an AI agent's effective authority by turning OpenAI's misalignment reports into staging tests with observable pass conditions and external evidence.
DeepSeek V4.1 Flash vs V4 Pro: What Actually Changed?
Choose DeepSeek V4.1 Flash or V4 Pro for coding, reasoning, writing, vision, and document workflows, with API requests, price examples, and migration checks.
GPT-6 Astra vs Sol vs Terra: What a Small Test Can—and Cannot—Tell You
Choose between GPT-6 Astra, GPT-5.6 Sol, and Terra using saved answers, scoring gaps, six practical task scenarios, and worked cost examples.
When to Use One AI Agent vs Multiple Agents
A practical decision guide for knowing when subagents create useful parallel work and when one agent will be faster, cheaper, more coherent, and easier to verify.
How to Test an AI Browser Agent Safely
Learn how to test a browser agent with isolated state, effect-based action boundaries, adversarial pages, evidence-matched assertions, and a release gate that preserves failures.
How to Use AI as a Math Tutor—Find the First Wrong Step Without Revealing the Answer
Use AI to find the first invalid algebra step, ask for one helpful hint, and check your revision. Includes worked error cases and an exact checker with explicit limits.
How to Test an AI Workflow for Indirect Prompt Injection
Test indirect prompt injection with a clean baseline, delivered hostile source, synthetic canary, independent action trace, and explicit pass, fail, and inconclusive decisions.
How to Build an AI Evaluation Set That Catches Regressions
Build a compact AI evaluation set with explicit labels, must-preserve details, forbidden claims, deterministic grading, and a release decision that exposes failures instead of averaging them away.
What Not to Upload to ChatGPT: An Artifact Checklist
Check credentials, private records, hidden file content, ChatGPT storage, and the next steps after a mistaken upload.
How to Protect Sensitive Data When Using AI
Choose when to proceed, sanitize locally, or stop before sharing sensitive data with AI; inspect the input, workspace, connectors, output, and release gate.
Also relevant
Related guides from other main topics
These overlap with AI Fundamentals, but their main subject belongs elsewhere.
Can One Agent Skill Work in Both Codex and Grok Build?
One SKILL.md can work in Codex and Grok Build. Download the exact Markdown-link checker used in both hosts, compare the complete files, and see which discovery and permission behaviors still need separate tests.
Why it fits here: The comparison tests which agent capability behavior transfers across hosts.
Qwen-Image-2.1 for Production Assets: What a 196-Image Test Supports
A production-focused reading of Li's 196-output Qwen-Image-2.1 report: what the examples support, where they fail, and which acceptance checks matter before shipping an asset.
Why it fits here: The analysis separates a third-party model report from firsthand testing.
Qwen3.8-Omni-Flash Video Analysis Test: Can It Find the Right Moment and Support Its Answer?
In one original short-video test, Qwen3.8-Omni-Flash found the saved review window and final status, refused an invented metric, and made one unsupported future-state inference.
Why it fits here: A bounded multimodal test separates observed model behavior from unsupported inference.
What Is Jev AI? How to Use TypeSafe's Decision Model
Jev answers predefined questions about text with typed choices, scores, or yes probabilities. Learn when that helps, how to make a first call, and how to check a routing workflow against your own cases.
Why it fits here: The guide explains a typed AI decision model and its limits.
How to Test GPT Image 2.5 Flare vs Sunburst
Compare Flare and Sunburst for your image API workflow: start with a six-request case, then test seven workload scenarios, quality, timing, failures, and accepted-image cost.
Why it fits here: The comparison defines what a bounded model test can show.
How to Find, Install, and Audit AI Agent Skills
A practical, source-bounded workflow for interpreting skills.sh rankings, reviewing Agent Skill files and permissions, choosing an installation path, and running an isolated first test.
Why it fits here: The intake workflow checks an agent capability before use.
How to Build Your First AI Agent—With Tools, Evals, and Human Approval
Build and inspect a Python agent with live model decisions, local tools, and exact human review. Includes custom-input and document-question walkthroughs, CLI and API adapters, and failure tests.
Why it fits here: The guide explains tools, evaluation, and human approval in a first agent.
How to Audit an MCP Server Before Connecting It to an AI Assistant
Follow an MCP server from the client entry to a limited test account, compare its actual tools and denied calls, and decide whether to connect it.
Why it fits here: The audit explains an agent integration trust boundary.
How to Add Human Approval Gates to AI Agent Tool Calls
Build a human approval gate around exact agent tool calls, with a runnable staging example, reviewer checks, changed-state protection, and failure handling.
Why it fits here: Approval boundaries are a general agent-control concept.
How to Create and Test a Codex Skill
Build a Codex Skill with a copyable entrypoint, a tested example, checks for other skill types, activation prompts, and fixes for common failures.
Why it fits here: The guide explains when and how an agent capability activates.
How to Write Unit Tests With ChatGPT That Catch Real Bugs
Write unit tests with ChatGPT from a clear behavior contract, run them, and check whether they fail for wrong results or unwanted service calls.