Research & Papers

The article explains the proposal of UniEvo-VL, a self-evolving framework where multimodal models learn from self-feedback, and the use of self-correction in test-time computation.

• •It leverages the ability of integrated multimodal systems for generation and understanding to learn from their own feedback.
• •It utilizes the model's own self-criticism as privileged information without relying on external teacher models.
• •It improves the model through constructive self-correction feedback during the test-time computation stage.

ELYZA launched an organization to advance recursive self-improvement (RSI) research and introduced initiatives for voice dialogue AI agents and physical AI using autonomous research foundations.

• •ELYZA established 'ELYZA RSI Research' to promote research and development in recursive self-improvement (RSI).
• •It introduced 'ELYZA Voice Agent,' a voice dialogue AI agent utilizing an autonomous research foundation.
• •Initiatives regarding physical AI were also announced.

An article sharing technical progress and achievements of AI in the field of mathematics.

An article examining the definition of "random" in random number generation and the problems that arise when it does not function as expected.

Explains the internal mechanisms of PyMC, a Bayesian statistics MCMC sampling library, and methods for visualizing models.

• •PyMC is an essential library for performing MCMC sampling in Bayesian statistics.
• •While previous discussions covered estimation and testing, this article focuses on PyMC's internal mechanisms.
• •It is published as a special feature (Episode 10) of the Bayesian statistics section of 'Easy Data Analysis for First-Year Professionals'.

コードが悪性かどうかをMoEモデルに尋ねた際、道徳性に関連するエキスパートパスが起動することが示された

• •コードの悪意判定をMoEモデルに問い合わせた際の動作が分析されている
• •モデルは道徳性に関連するエキスパートパスを選択的に使用することが確認された
• •LLMの内部メカニズムにおける倫理判断のプロセスに関する考察が含まれている

Google Researchが、地球規模のジオスペーシャル基盤モデルを公衆衛生分野に応用する研究について紹介している

Itoki and Matsuo Laboratory built an AI model analyzing behavioral data from 854 people to evaluate the impact of meetings and breaks on well-being and productivity.

• •An AI model was built by analyzing behavioral data from 854 employees.
• •Metrics include meeting duration and frequency, email and chat volume, and time spent in break areas.
• •The study explores how these behaviors affect employee productivity, physical and mental health, and interpersonal relationships.

ELYZA's CEO explains the 'recursive self-improvement' technology behind the monthly release of new frontier AI models and the current state of AI developing AI.

• •The frequent emergence of new models is driven by the technology of 'recursive self-improvement' (RSI).
• •It is noted that 'AI developing AI' has become a reality.
• •Soneoka, CEO of ELYZA under KDDI, discusses the current state of this technology.

A two-year randomized trial in 18 Tennessee schools provided large-scale experimental evidence on the educational impact of the AI tutor Khanmigo.

• •A cluster randomized trial was conducted over two years in 18 middle schools in Tennessee.
• •Students used Khanmigo, configured to coach rather than provide answers, during their existing daily remedial math sessions.
• •This study provides part of the large-scale experimental evidence as generative AI is expected to transform education.