AI · Official
New Research Aims to Prove an AI Response Really Came From the Claimed Model
Anthropic Research (mirror) ·
At a glance
New research shows it may be possible to confirm, using cryptographic proofs, that an AI model's answer really came from the model it claims to be. The work centers on a system called FlashProve, and the team describes it as an interim research update. They have also released related materials under an open license.
Key points
- Right now, users have a hard time checking whether an AI answer really came from the promised model.
- Cryptographic proofs (mathematical methods that show something is true without revealing secrets) could prove that an answer was generated by a specific model.
- The team says it created proofs at about 110 tokens per second for a large model on a single H100 GPU.
- Creating a proof takes about 70 to 95 times as long as simply running the model, according to the team.
Why it matters
If there is a way to verify that AI answers are trustworthy, it could help stop providers from secretly using cheaper models to save money.
Original source
Progress on provable inference for large language models
Anthropic Research (mirror) · Nov 2, 09:00 KST
Read on Anthropic Research (mirror) ↗AI summarized the original article. Check the original for accuracy. Copyright belongs to Anthropic Research (mirror).
More from this field
- AWS Adds Shared Cache Skill to Cut Self-Hosted LLM Cost and Latency
AWS wrote about a shared cache that lets self-hosted AI models reuse saved work across servers, aiming to cut cost and wait time. The post shows how a coding agent can help set it up. Amazon says it can cut inference cost by up to 40% and first-token delay by up to 80%.
Amazon AWS Machine Learning blog · Nov 21, 09:14 KST - Can AI Really Find Black Friday TV Deals? A BBC Test Says Not Yet
A BBC team tested four popular AI chatbots to see if they could find a good Black Friday deal on a 55-inch TV. The results were mixed, with many links broken or out of date.
BBC News · Nov 21, 09:06 KST - Anthropic Mathematician Says Claude Helped Crack the Riemann Hypothesis
A video about a claim that an Anthropic mathematician, Levent Alpöge, used Claude to find a proof of the Riemann Hypothesis in about six weeks. The creator skips the heavy math and looks at what this could mean for AI and its future. The proof is still being checked.
TheAIGRID · Nov 21, 08:45 KST - Google Brings Duplex Calling to Gemini, Letting AI Phone Businesses for You
Google has moved its Duplex phone-calling technology into the Gemini app, so Gemini can now call businesses for you. The system has changed a lot since 2018, shifting from narrow custom models to a single stack built on Gemini.
Google Research blog · Nov 21, 08:30 KST