Causal tracing is an emerging technique in LLM observability that attempts to identify which internal components, such as neurons or attention heads, were most influential in producing a specific output.

Authors

• Person: Not available Organization: TTMS
  LLM Observability: How to Monitor AI When It Thinks in Tokens | TTMS

Sources

• LLM Observability: How to Monitor AI When It Thinks in Tokens | TTMS ttms.com TTMS via serper

Referenced by nodes (2)

• neurons concept
• LLM observability concept