MAROKO133 Hot ai: MIT Researchers Unveil “SEAL”: A New Step Towards Self-Improving AI Hari

📌 MAROKO133 Breaking ai: MIT Researchers Unveil “SEAL”: A New Step Towards Self-Im

The concept of AI self-improvement has been a hot topic in recent research circles, with a flurry of papers emerging and prominent figures like OpenAI CEO Sam Altman weighing in on the future of self-evolving intelligent systems. Now, a new paper from MIT, titled “Self-Adapting Language Models,” introduces SEAL (Self-Adapting LLMs), a novel framework that allows large language models (LLMs) to update their own weights. This development is seen as another significant step towards the realization of truly self-evolving AI.

The research paper, published yesterday, has already ignited considerable discussion, including on Hacker News. SEAL proposes a method where an LLM can generate its own training data through “self-editing” and subsequently update its weights based on new inputs. Crucially, this self-editing process is learned via reinforcement learning, with the reward mechanism tied to the updated model’s downstream performance.

The timing of this paper is particularly notable given the recent surge in interest surrounding AI self-evolution. Earlier this month, several other research efforts garnered attention, including Sakana AI and the University of British Columbia’s “Darwin-Gödel Machine (DGM),” CMU’s “Self-Rewarding Training (SRT),” Shanghai Jiao Tong University’s “MM-UPT” framework for continuous self-improvement in multimodal large models, and the “UI-Genie” self-improvement framework from The Chinese University of Hong Kong in collaboration with vivo.

Adding to the buzz, OpenAI CEO Sam Altman recently shared his vision of a future with self-improving AI and robots in his blog post, “The Gentle Singularity.” He posited that while the initial millions of humanoid robots would need traditional manufacturing, they would then be able to “operate the entire supply chain to build more robots, which can in turn build more chip fabrication facilities, data centers, and so on.” This was quickly followed by a tweet from @VraserX, claiming an OpenAI insider revealed the company was already running recursively self-improving AI internally, a claim that sparked widespread debate about its veracity.

Regardless of the specifics of internal OpenAI developments, the MIT paper on SEAL provides concrete evidence of AI’s progression towards self-evolution.

Understanding SEAL: Self-Adapting Language Models

The core idea behind SEAL is to enable language models to improve themselves when encountering new data by generating their own synthetic data and optimizing their parameters through self-editing. The model’s training objective is to directly generate these self-edits (SEs) using data provided within the model’s context.

The generation of these self-edits is learned through reinforcement learning. The model is rewarded when the generated self-edits, once applied, lead to improved performance on the target task. Therefore, SEAL can be conceptualized as an algorithm with two nested loops: an outer reinforcement learning (RL) loop that optimizes the generation of self-edits, and an inner update loop that uses the generated self-edits to update the model via gradient descent.

This method can be viewed as an instance of meta-learning, where the focus is on how to generate effective self-edits in a meta-learning fashion.

A General Framework

SEAL operates on a single task instance (C,τ), where C is context information relevant to the task, and τ defines the downstream evaluation for assessing the model’s adaptation. For example, in a knowledge integration task, C might be a passage to be integrated into the model’s internal knowledge, and τ a set of questions about that passage.

Given C, the model generates a self-edit SE, which then updates its parameters through supervised fine-tuning: θ′←SFT(θ,SE). Reinforcement learning is used to optimize this self-edit generation: the model performs an action (generates SE), receives a reward r based on LMθ′’s performance on τ, and updates its policy to maximize the expected reward.

The researchers found that traditional online policy methods like GRPO and PPO led to unstable training. They ultimately opted for ReST^EM, a simpler, filtering-based behavioral cloning approach from a DeepMind paper. This method can be viewed as an Expectation-Maximization (EM) process, where the E-step samples candidate outputs from the current model policy, and the M-step reinforces only those samples that yield a positive reward through supervised fine-tuning.

The paper also notes that while the current implementation uses a single model to generate and learn from self-edits, these roles could be separated in a “teacher-student” setup.

Instantiating SEAL in Specific Domains

The MIT team instantiated SEAL in two specific domains: knowledge integration and few-shot learning.

• Knowledge Integration: The goal here is to effectively integrate information from articles into the model’s weights.
• Few-Shot Learning: This involves the model adapting to new tasks with very few examples.

Experimental Results

The experimental results for both few-shot learning and knowledge integration demonstrate the effectiveness of the SEAL framework.

In few-shot learning, using a Llama-3.2-1B-Instruct model, SEAL significantly improved adaptation success rates, achieving 72.5% compared to 20% for models using basic self-edits without RL training, and 0% without adaptation. While still below “Oracle TTT” (an idealized baseline), this indicates substantial progress.

For knowledge integration, using a larger Qwen2.5-7B model to integrate new facts from SQuAD articles, SEAL consistently outperformed baseline methods. Training with synthetically generated data from the base Qwen-2.5-7B model already showed notable improvements, and subsequent reinforcement learning further boosted performance. The accuracy also showed rapid improvement over external RL iterations, often surpassing setups using GPT-4.1 generated data within just two iterations.

Qualitative examples from the paper illustrate how reinforcement learning leads to the generation of more detailed self-edits, resulting in improved performance.

While promising, the researchers also acknowledge some limitations of the SEAL framework, including aspects related to catastrophic forgetting, computational overhead, and context-dependent evaluation. These are discussed in detail in the original paper.

Original Paper: https://arxiv.org/pdf/2506.10943

Project Site: https://jyopari.github.io/posts/seal

Github Repo: https://github.com/Continual-Intelligence/SEAL

The post MIT Researchers Unveil “SEAL”: A New Step Towards Self-Improving AI first appeared on Synced.

🔗 Sumber: syncedreview.com

📌 MAROKO133 Update ai: China building new 7.2 GW nuclear plant to power energy nee

China General Nuclear Power Group (CGN) has officially commenced full-scale construction on a massive new energy facility in East China’s Shandong Province, a project designed to generate 50 billion kilowatt-hours of electricity annually. 

Once completed, the Zhaoyuan Nuclear Power Plant will produce enough power to meet the household needs of approximately 5 million people.

The project serves as a significant environmental counterweight, with the full facility expected to cut standard coal consumption by 15.27 million tons and reduce carbon dioxide emissions by 46.2 million tons each year. 

According to project planners, this reduction offers an environmental benefit equivalent to planting more than 110,000 hectares of forest. The site is ultimately planned to house six Hualong One reactors, totaling an installed capacity of about 7.2 million kilowatts (7.2 GWe).

Cooling tower debut with innovative design

Construction on Unit 1 began Tuesday with the pouring of the first concrete, marking the debut of a 203-meter-high (666-foot-high) natural draft cooling tower—a first for the Hualong One reactor model. 

With a spray area of 16,800 square meters, this massive structure enables the plant to shift its primary cooling source from the ocean to the atmosphere. Yu Xiangdong, executive director of Shandong Zhaoyuan Nuclear Power Co, noted that this innovation reduces energy consumption and allows for the reuse of water resources.

The introduction of the cooling tower represents a strategic pivot in design. While coastal plants typically rely on direct seawater cooling, this new “secondary-circuit cooling technology” utilizes an atmospheric heat-sink effect. 

“By adopting this technology, the Zhaoyuan project expands the siting possibilities for nuclear facilities and provides new reference experience for China’s future nuclear development,” Yu said, as reported by Global Times.

The facility also introduces a “natural + mechanical” dual-layer safety system designed to provide robust operational buffers. Yang Yazhang, deputy general manager of CGN Power Engineering Co, explained that the natural draft tower can sustain operations for at least two hours if the external water supply is lost, ensuring sufficient time for a safe shutdown. 

Complementing this, the plant is the first Hualong One unit to feature a nuclear-grade mechanical draft tower equipped with a large on-site water tank capable of sustaining reactor cooling for at least 30 days without replenishment.

“These two systems together form a ‘natural + mechanical’ secondary-circuit cooling configuration, creating a comprehensive, dual-layer cooling-source protection system that covers both the conventional and nuclear islands,” Yang added.

Industry versatility for future development

Lin Boqiang, director of the China Center for Energy Economics Research at Xiamen University, highlighted the significance of adapting this technology for a coastal site.

While cooling towers are standard for inland projects, Lin noted that integrating them with Hualong One technology demonstrates the versatility of China’s manufacturing base. 

He added that the project leaves the industry well-positioned to support a sustainable energy transformation, proving that nuclear facilities can be adapted to meet specific environmental requirements or resource limitations.

🔗 Sumber: interestingengineering.com