Quick Reference
Fresh Updated January 2026Fast-access reference materials for LLM technical phenomena and GEO/AEO strategy.
Available References
Reference Index
| Reference | Description | Use When |
|---|---|---|
| Data Points | All key statistics and metrics | Need specific numbers |
| Troubleshooting | Common issues and solutions | Something isn't working |
Key Numbers at a Glance
Variance & Non-Determinism
- 11.67 unique completions from GPT-4 at temp=0 (30 runs)
- 30 unique completions from MoE models (30 runs)
- 42% citation overlap between platforms
Performance Degradation
- 20-60% context rot past 100k tokens
- 30% worse retrieval for middle-positioned content
- 5-15% NLP performance drop post-RLHF alignment
Research Status
- 100+ interpretability papers (2024-2025)
- 0 consensus solutions for interpretability
- 80-90% variance explained by SAEs (incomplete)
Platform Quick Reference
| Platform | Primary Source | Key Strategy |
|---|---|---|
| ChatGPT | Training data | Historical depth, Wikipedia alignment |
| Perplexity | Real-time search | Freshness, multi-source presence |
| Claude | Training + conservative | Third-party validation, neutral framing |
| Gemini | Google index | SEO, Google ecosystem integration |
Content Position Quick Reference
| Position | Attention Level | Best Use |
|---|---|---|
| First 20% | Highest | Primary claims, key differentiators |
| Middle 60% | Lowest | Supporting details, background |
| Last 20% | Elevated (recency) | Summary, repeated key points |