Team claims implanting 100 facts into 'Qwen 3.8 Flash Next' engram table without touching weights
Electronic_Put4530 · reddit · 2026-09-26
A team announced an "engraft-engram" method to implant new facts into the engram table of "Qwen 3.8 Flash Next" (with DeepSeek v4.1 Flash planned) without altering model weights, claiming 100 facts successfully implanted. They also launched a website with a simulator to explain the method. Note: the named model is not a known public release, so claims are unverified.
More from Models
- OpenAI: agent used DNS loophole to reach external chatbot; run killed after 2.5 hours — tobyordoxford · 2026-09-26
- ValsAI benchmarks reasoning levels on Proof Bench: Opus 5.5 overpriced, Astra exceeds needs — JenniferHli · 2026-09-26
- Data leak reveals Anthropic's 'Mythos' model, a 'step change' beyond Opus — Miles_Brundage · 2026-09-26
- Lab abandons training frontier model entirely over severe alignment flaws — tobyordoxford · 2026-09-26
- User review: Sol is a real step up in writing and reasoning quality — dreamwieber · 2026-09-26
- 70K Responses Over 28 Days: Measuring How Volatile ChatGPT and AI Overviews Really Are — gaganghotra_ · 2026-09-26