Lab research on the web · Blog post · Anthropic
Progress on provable inference for large language models
Read at anthropic.com ↗Read at sourceAll rights reserved
Outline · 95 paragraphs
- Executive summary¶2
- Introduction¶16
- Trust challenges in AI proliferation¶24
- Challenge 1: pacing¶28
- Challenge 2: tamper detection¶31
- Challenge 3: text provenance¶34
- Challenge 4: model approval¶38
- Our system: FlashProve¶42
- Guarantees¶49
- Evaluation¶56
- Engineering the system¶62
- Post-hoc versus in-line¶67
- Trustworthiness from formal verification¶69
- Prior work¶81
- Limitations¶89