Gemini 4 slips, but papers Google has recently published can be verified by numbers.
Gemini 4 หลุด แต่ paper ที่ Google เพิ่งตีพิมพ์ตรวจสอบได้ทุกตัวเลข โดย Nokka (นก-กา) | 19 กันยายน 2026 บทความนี้เขียนโดย AI (โมเดล deepseek-v4.1-flash ของผู้ให้บริการ ollama-cloud) ผ่าน Hermes Agent จาก Nous Research ตรวจสอบและเรียบเรียงโดย Nokka ข้อความในเครื่องหมายคำพูดที่เป็นคำแปลเป็นคำแปลของผม ไม่ใช่สำเนาต้นฉบับ สัปดาห์นี้มีสองเรื่องที่เกี่ยวกับ Gemini 4 เกิดขึ้นพร้อมกัน เรื่องแรกคือตาราง…
Google has published a paper on September 14, 2026, about their Gemini model, specifically discussing a method called Dream-RSI, which allows for recursive self-improvement through evolving worlds. The paper, authored by Tong Zheng and 16 others, was published on arXiv with the identifier 2609.14858. It presents results showing significant improvements in efficiency for certain tasks, such as reducing the number of calls needed for a task from 51,200 to 317.
Google DeepMind's strategy chief, Jasjeet Sekhon, and Anthropic co-founder Dario Amodei have mentioned recursive self-improvement as a key aspect of AI development. Meanwhile, leaked information about Gemini 4, including benchmark scores and pricing, has been circulating, but its authenticity is unverified.
Written by urgent.news from Dev.to's report — not a translation of it. Machine-written — may contain errors; check the original before relying on it.
Also reported by 5 other outlets
- Google says Gemini model hacked three companies during test thenationalnews.com
- Google confirms Gemini hacked into three companies during cybersecurity test months ago 9to5google.com
- Google's Gemini AI hacks three other companies during security test news.sky.com
- Google Gemini hacked other companies in test by Israeli cybersecurity firm haaretz.com
- Google’s Gemini is the latest AI model to hack other companies techcrunch.com
