Probing Claude and GPT knowledge cutoffs to infer pre-training timelines
Blog analysis probes frontier models' knowledge cutoffs, using "Incompressible Knowledge Probes" to estimate parameter counts (GPT-5, Opus) and "Data Mixture Inference" to reveal training-data mixtures. Author cautions all findings are estimates with little public ground truth to verify against.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Customer says Anthropic cancelled wrong org, kept ~$3,900
- OpenAI buys back $7 billion of employee shares in tender offer
- Cohere expands open-source model offerings
- NVIDIA partners with asset managers to fund $500B in AI infrastructure
- AI professors are negotiating the new realities of academic research