Latest AI Trends: Book Extraction from LLMs, Psychedelics & Creativity, Autonomous Driving Race
Tak@Here are today's top AI & Tech news picks, curated with professional analysis.
プロダクション言語モデルからの書籍抽出：著作権リスクの検証
Expert Analysis
This paper investigates the feasibility of extracting copyrighted text, specifically entire books, from large production language models (LLMs) like Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, and Grok 3. The researchers developed a two-phase procedure involving initial probes (sometimes using a 'Best-of-N' jailbreak) and iterative continuation prompts. They found that even with system-level safeguards, significant amounts of text could be extracted from these models. In some cases, jailbroken models output entire books with high verbatim recall. The study highlights that extraction of in-copyright training data remains a risk for production LLMs, despite implemented safeguards.
👉 Read the full article on arXiv
- Key Takeaway: Production LLMs, even with safety measures, are vulnerable to extracting copyrighted training data, posing a significant legal and ethical risk.
- Author: Ahmed Ahmed, A. Feder Cooper, Sanmi Koyejo, Percy Liang
過去のサイケデリック使用歴が発散的思考を予測する
Expert Analysis
This study, involving 5,905 participants from the Great British Intelligence Test, explored the relationship between past psychedelic use and creativity. The researchers compared individuals with and without a history of psychedelic use across three components of creativity: divergent thinking, cognitive reflection, and insight. Results indicated that psychedelic users scored significantly higher on divergent thinking compared to both non-drug users and drug users who had not used psychedelics. However, no significant differences were found in cognitive reflection, number of insights, or insight accuracy. The study suggests a potential association between naturalistic psychedelic use and enhanced divergent thinking, while cautioning that causality needs to be established through prospective and controlled studies.
👉 Read the full article on arXiv
- Key Takeaway: Naturalistic psychedelic use is associated with enhanced divergent thinking, a key component of creativity, but not necessarily with improved insight or cognitive reflection.
- Author: Gregory J Pope, Christopher Timmermann, William Trender, Peter J Hellyer, Maria Bălăeţ, Ruben E. Laukkonen
NvidiaのTesla FSD競合をテスト：Teslaは警戒すべき
Expert Analysis
This article reviews Nvidia's autonomous driving technology, showcased at CES 2026, which is positioned as a competitor to Tesla's Full Self-Driving (FSD). Nvidia's Alpamayo AI platform utilizes sensor fusion (lidar, radar, cameras) and aims to enable vehicles to 'understand, reason and act.' Early demonstrations with partners like Mercedes-Benz showed impressive capabilities, rivaling Tesla's FSD in complex scenarios. However, the article notes that Tesla's advantage lies in its vast real-world driving data and its ongoing efforts to address edge cases. Nvidia's strategy is to supply automakers with hardware and software, leveraging an open-source approach to foster industry-wide innovation. While Nvidia's technology is advanced, Tesla's extensive real-world testing and data collection provide a significant competitive edge.
👉 Read the full article on The Verge
- Key Takeaway: Nvidia presents a strong technological challenge to Tesla's FSD with its sensor-fusion approach and reasoning capabilities, but Tesla's real-world data advantage remains a critical differentiator.
- Author: Andrew J. Hawkins
Follow me!
