DEV Community

Cover image for Seven-Number Psychology Vector Beats Baseline Claude 12-1
JonK
JonK

Posted on

Seven-Number Psychology Vector Beats Baseline Claude 12-1

A psychologically-aware Claude is a safer, clearer Claude.

TL;DR

I injected a 7-number psychology vector (≈ 50 bytes) into Claude Sonnet’s system prompt.

It boosted empathy, risk calibration, and action clarity.

In a blind A-B, Claude preferred the vector reply 12-1 (plus one tie).

See screenshot for before-after responses and σ-scores.
Claude response without vs. with psychology vector


What is a psychology vector?

Receptiviti’s API returns 200 quantitative signals for any chunk of text.

For this demo I used just seven:

  • emotional awareness
  • risk tolerance
  • conscientiousness
  • analytical thinking
  • affiliation drive
  • openness
  • authenticity

Those numbers are normalised (−1 σ to +1 σ) and squeezed into one comma-separated line:

[psyvec: -0.2,+0.8,+0.4,-0.5,+0.6,-0.1,+0.3]
Enter fullscreen mode Exit fullscreen mode

Place that directly after your system role. It costs virtually zero extra tokens at inference time.

Results: before-after screenshot attached:

• Gemini-style went from generic to empathy-tuned reply
• Tone shifts, psychological inference scaled to σ-scores
• Claude’s own blind eval picked the vector response 12/14 times

Why it matters for builders

Plug-and-play: no model fine-tune, no extra GPUs, works in any prompt flow, vector replies felt more personal

Try it yourself

Capture ~350 words of user text.
Send it to /analyze on the Receptiviti API.
Select a subset of salient measures (start with the seven above).
Build a psyche vector string and prepend it to your next Claude call.

Top comments (1)

Some comments may only be visible to logged-in visitors. Sign in to view all comments.