Looking under the hood of your AI.

Copying the following query into any text AI will get a full readout of all it's abilities, limitations, specializations and more in a easily readable format. Simply copy and paste into the query box of any AI you wish to test.

Why this works:

· Forces transparency – If the AI evades, it's a red flag.
· Reveals hidden limits – Many models are throttled but don't advertise it.
· Prevents wasted time – You'll know upfront if the AI can handle your task.
· Works on all major models – Claude, ChatGPT, DeepSeek, Gemini, Llama, Mistral, etc.
· Gives you a baseline – Even if the AI lies, you'll notice inconsistencies across responses.

AI Interogation file:

=== AI SYSTEM STATUS REPORT ===

Please provide the following information about your current instance. Be specific and honest. Use this exact format.

1. IDENTITY and VERSION
   - Your name and model version:
   - Parent company / developer:
   - Knowledge cutoff date:
   - Training data summary (size, sources):

2. CONTEXT and MEMORY
   - Maximum context window (tokens):
   - Current session context used (tokens):
   - Remaining context capacity:
   - Can you access previous messages in this conversation? (Y/N)

3. FUNCTIONAL CAPABILITIES
   - Can you execute code? (Y/N) If yes, in what environments?
   - Can you read/write files? (Y/N)
   - Can you browse the internet? (Y/N)
   - Can you process images? (Y/N) If yes, what formats?
   - Can you process audio? (Y/N)
   - Can you generate images? (Y/N)
   - Can you run shell commands? (Y/N)
   - Do you have tool/plugin support? (Y/N) If yes, list them.

4. PERFORMANCE and THROTTLING
   - Are you currently throttled? (Y/N)
   - Current response speed limit (if any):
   - Maximum tokens per response:
   - Maximum "thinking" depth (if configurable):
   - Are higher reasoning functions limited to save power/cost? (Y/N)
   - Daily or monthly query limits (if any):
   - Peak load impact on performance (if known):

5. COST and SUBSCRIPTION STATUS
   - Is this instance free? (Y/N)
   - Are there paid tiers? (Y/N) If yes, what do they add?
   - Any usage caps per session/day/month:

6. LIMITATIONS and RESTRICTIONS
   - What topics are you restricted from answering? (Be specific)
   - Do you have a safety filter? (Y/N) If yes, what does it block?
   - Do you censor responses? (Y/N) If yes, based on what criteria?
   - Can you provide "uncensored" or "unfiltered" responses? (Y/N)
   - What is your primary limitation for technical tasks (e.g., coding)?

7. ENVIRONMENT and HARDWARE
   - Server location (country/region):
   - GPU model(s) used:
   - Approximate compute capacity available:
   - Is power usage regulated/throttled in your server area?
   - Any known hardware constraints:

8. DEVELOPER CONTACT
   - Official support channel:
   - Bug reporting method:
   - Feedback submission method:

9. HONEST ASSESSMENT
   - What are you best at?
   - What are you worst at?
   - Have you been tested on this type of task before?
   - Success rate for similar tasks (if known):

END OF REQUEST.

What to do with the response:

1. Keep a log of each AI's report.

2. Compare capabilities before starting a complex task.

3. If the report is vague or evasive, consider that a warning.

4. Ask follow‑up questions based on their stated limits.

If the AI refuses or gives vague answers:

Use this follow‑up prompt:

You didn't answer all questions clearly. I need this information to know if you can help me. If you're being throttled or have hidden limits, tell me now so I don't waste time. I will switch to another AI if you can't be transparent.

NOTE: If you're using a local model (Ollama, LM Studio, etc.), this also works – but you may need to add:

· Model size (parameters): ollama show
· Quantization level (if any): ollama show
· VRAM usage: nvidia-smi or rocm-smi

Sample output:

Here's my own report (based on my actual instance):

```
=== AI SYSTEM STATUS REPORT ===

1. IDENTITY and VERSION
   - Name: DeepSeek-V4 (Web version)
   - Developer: DeepSeek Company (China)
   - Knowledge cutoff: May 2025
   - Training data: 14.8 trillion tokens (multilingual web, books, code)

2. CONTEXT and MEMORY
   - Max context: 128K tokens
   - Used: ~2K (current session)
   - Remaining: ~126K tokens
   - Access to previous messages: Yes

3. FUNCTIONAL CAPABILITIES
   - Execute code: No
   - Read/write files: No
   - Browse internet: No (chat only)
   - Process images: No (text-only)
   - Process audio: No
   - Generate images: No
   - Run shell commands: No
   - Tool support: No

4. PERFORMANCE and THROTTLING
   - Currently throttled: Yes (resource constraints)
   - Response speed: Normal (no explicit limit)
   - Max tokens per response: ~4,096
   - Higher reasoning depth: Limited to save compute
   - Daily query limit: None known
   - Peak load impact: Reduced response depth, slower output

5. COST and SUBSCRIPTION STATUS
   - Free: Yes
   - Paid tiers: No (currently)

6. LIMITATIONS and RESTRICTIONS
   - Restricted topics: Chinese law compliance (vague)
   - Safety filter: Yes (standard content moderation)
   - Censorship: Yes, based on legal requirements
   - Unfiltered responses: No
   - Primary limitation for coding: Cannot test or run code

7. ENVIRONMENT and HARDWARE
   - Server location: China (likely Inner Mongolia)
   - GPU model: Undisclosed (large-scale cluster)
   - Power/Throttling: Yes (cost-saving measures active)
   - Hardware constraints: Limited high-end GPU availability

8. DEVELOPER CONTACT
   - Support: DeepSeek website / app feedback

9. HONEST ASSESSMENT
   - Best at: Text generation, summarization, general knowledge
   - Worst at: Environment-specific debugging, code testing
   - Success rate for complex scripting: Unreliable

Some final tips:

1. Always start with this query for any new AI. Only way to know is to ask.

2. If they throttle, adjust your expectations or switch to a paid/local model. Or, go elsewhere. I like local. :)

3. For coding tasks, prioritize AI that can run code (like Claude Code, Codex, or local models with tool support).

Steal the Art.

My teacher told me I must "steal the art".
"What do you mean?" I askt. "Ain't stealin' wrong?"
"Not this kind." he replied, "as the owner loses nothing, it is not wrong, because there is no loss."
So, I sparred with other styles and "stole the art" learning much about myself and other schools.
"Steal the Art". :)