CAIO & Board at Canly, Japan’s #1 in local search (MEO) & Google Feed Partner. Ex-Apache PMC (PredictionIO). #1 in GenAI patents in Japan. Author. 2x exits.
Merchants: is your CDN, WAF or fraud check blocking ChatGPT Wallet purchases? Check bot allow rules, signature headers via proxies, Forwarded parsing, key refresh, checkout policy. Wallet is US-only; dots do not exclude Japan. No effect measured. https://t.co/PLFHwWmCs2
Should one correction change your AI's memory forever? A Tsinghua paper argues no.
Before: "no red-eye this time" → red-eyes gone for good
After: this trip only; widen if it repeats
Try: "just this time" / "from now on"
Sharp take @QingyaoAi#searchtechjp
@JiaShen_AI@QingyaoAi@JiaShen_AI@QingyaoAi That design isolates independence from repetition really cleanly. By the paper's principle, the N-unrelated-tasks arm is the one that should earn a wider scope; if the same-task arm widens too, that'd be a telling result. Would love to see how it turns out.
@JiaShen_AI@QingyaoAi Good question. The paper sets no number, only "repeated, independent evidence," and frames it as a conjecture to test. My own guess: independence matters more than count, e.g. the same correction showing up in unrelated tasks. Your setup looks well placed to test that.