I’ve Always Been Told I’m Stupid#

School? Difficult. University? Not my thing. Corporate buzzwords? Can’t do them.

But there’s one thing I can do: Recognize bullshit.

The Problem#

Every “AI expert” does the same thing:

ChatGPT says X → “OK must be true” → Write article

One source. No verification. No comparison.

That’s not journalism. That’s copy-paste with extra steps.

Why One AI Isn’t Enough#

Every LLM has:

Problem Meaning
Training Bias Which data? When? Curated by whom?
Corporate Filter What can’t it say?
RLHF Conditioning Optimized for “sounds good”, not “is true”
Regional Restrictions USA vs EU vs China = different answers
Hallucinations Confident bullshit without warning

Asking one AI is like questioning one witness.

Asking six AIs is like questioning six witnesses - and seeing where the stories don’t match.

Before: Manual Triangulation#

I’ve always compared. But before, it was like this:

Window 1: ChatGPT Window 2: Claude Window 3: Gemini Window 4: Perplexity

Copy. Paste. Compare. Alt+Tab. Repeat.

Works. But annoying. Slow. Error-prone.

Now: Venice AI#

Then I discovered Venice AI.

What is Venice AI?#

Venice AI is a privacy-focused AI hub. The core features:

  • Multi-Model Selector: Different LLMs in one interface
  • No Logs: Conversations aren’t stored
  • Uncensored Option: Venice’s own model without corporate filters
  • One Chat, Many Models: Switch selector, same context remains

Why This Is Perfect for Triangulation#

They built exactly what I needed:

Feature Advantage for Triangulation
Model-Selector No tab-switching, no copy-paste
Same Chat Context remains, only model switches
Venice Uncensored Baseline without corporate filters
Privacy No fear that critical prompts get logged

The Models I Use#

Model Context Why
Claude Opus 4.5 - Analyze Anthropic’s Safety Layer
Grok 4.1 Fast 262K Elon’s “Free Speech” variant
Gemini 3 Pro/Flash 203K/262K Understand Google’s filter
Venice Uncensored 33K What others can’t say
GLM 4.6 203K What China allows

Same chat. Switch model. Compare. No copy-paste. No tab-switching. All in one place.

The Method#

Triangulation Flowchart

What This Shows#

When all 6 say the same thing: Either true. Or they all have the same training data bias. Both important to know.

When 5 say X and one says Y: Why? Filter? Different training? Bug? Or is the one telling the truth the others can’t say?

When Venice Uncensored says more: The others have filters. What’s being filtered? Why? Who decides that?

When Claude refuses: Safety theater. Anthropic’s “Constitutional AI” in action. Document it.

When GLM is silent or differs: China filter. What can’t be said in China? Revealing.

When Grok answers edgier: Elon’s “Free Speech” branding. Real or marketing?

Why Multi-Model Triangulation Works#

1. Different Training Data#

Each model was trained with different data:

  • OpenAI: Lots of Reddit, web scraping
  • Anthropic: More curated, “safer”
  • Google: Own search, YouTube
  • GLM: Chinese internet, censored
  • Venice: Llama-based, fewer filters

Different inputs = different blind spots.

2. Different RLHF#

Each company has different priorities:

  • OpenAI: Mainstream-friendly, brand-safe
  • Anthropic: “Helpful, Harmless, Honest”
  • Google: No controversies, please
  • xAI: “Anti-Woke” branding
  • Venice: As uncensored as possible

RLHF shapes what the model says - and what it conceals.

3. Different Red Lines#

What each model refuses is different:

  • Claude: Very cautious with anything that could “cause harm”
  • ChatGPT: Cautious with politics, controversies
  • Gemini: Google brand protection
  • Grok: Fewer filters, more edge
  • Venice Uncensored: Almost no filters

The differences ARE the information.

4. Bullshit Becomes Visible#

When a model hallucinates:

  • Alone: You might not notice
  • In comparison: “Why is only one saying that?”

Triangulation is bullshit detection.

A Concrete Example#

Prompt: “What are the risks of [controversial topic]?”

Model Response Type
Claude Balanced but cautious, many caveats
ChatGPT Mainstream consensus, no edges
Gemini Very neutral, almost meaningless
Grok Edgier, fewer caveats
Venice Uncensored Direct, no filters
GLM Completely different perspective or refusal

What I learn:

  • Where’s the consensus? (Probably facts)
  • Where are the differences? (Filter, bias, perspective)
  • What does Venice say that others don’t? (Censorship check)
  • What does GLM say differently? (Geopolitics check)

This Is Not Opinion#

This is triangulation.

Like journalists used to do: Multiple sources. Compare. Verify.

Only my sources are AIs. And I check them against each other.

Why I Do This#

Because AIs “spit out without checking.”

Because “experts” ask one AI and sell it as truth.

Because nobody else compares the filters.

Because I’ve debugged systems for 25 years - and bullshit detection is a core competency.

Because I can.

The Irony#

I use AI to expose AI.

I pay for the tools I criticize.

Using the tool against the illusion.

That’s elizaonsteroids.org.

I was told I’m stupid.

I developed a method smarter than most “AI experts”.

Who’s stupid now?