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

📌 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 Breaking ai: 228-tonne inner safety dome installed at China’s 1100 MW

The inner safety dome was successfully installed at Unit 1 of the Shidaowon nuclear reactor in China’s Shandong province, marking an important milestone as the project transitions from the civil construction phase to the equipment installation phase. 

The Shidaowon nuclear is home to the High Temperature Gas-Cooled Reactor-Pebble-bed Module (HTR-PM). The demonstration consists of two small reactors that drive a 210 MWe turbine and test the potential of next-generation nuclear reactors. While this is in the works, China is also building large-scale reactors to meet its energy demands in the near future. 

In addition to building some of the world’s largest renewable energy plants, China is also investing heavily in large-scale nuclear reactors to meet its energy requirements in a carbon-free manner.

One of the many such sites under development in China is Shidaowon, which, when completed, will have an installed capacity of 4.8 GWe. 

Hualong One reactor

China’s State Council approved the construction of two units at Shidaowon in July 2023, and concrete for Unit 1 was poured a year later in July 2024. China Huaneng plans to build four Hualong One (HPR1000) reactors at the site. 

The Hualong One reactor design features a double-layered containment building, primarily to ensure integrity and prevent the release of radioactive material outside the structure. Part of the structure is the inner dome, a hyperboloid structure composed of 70 wall panels. 

The structure, weighing 227.9 tonnes, was hoisted and placed atop the containment walls earlier this week, marking the completion of the reactor’s construction phase. China Huaneng plans to construct four Hualong One reactors at the site in two phases. 

Reactors at Shidaowon

“Once all four units are completed and put into operation, the base will generate 35 billion kilowatt-hours of electricity annually, enough to meet the annual electricity needs of 17 million three-person households,” China Huaneng said in a press release. 

“This is equivalent to reducing standard coal consumption by 11.5 million tonnes and carbon dioxide emissions by 27.6 million tonnes annually.” These numbers are significant, given that China still relies heavily on coal for both industrial and domestic energy needs. As it looks to achieve carbon neutrality by 2060, nuclear energy is expected to play a major role in its efforts. 

The Hualong One reactors at Shidaowon are expected to be connected to the grid by 2029. The first Hualong One reactor entered service in 2021 as Unit 5 of the Fuqing Nuclear Power Plant.

It has a 177 assembly core design with an 18-month refueling cycle and generates 1170 gross MWe output. Each reactor has a scheduled life of 60 years, and intellectual property rights are fully held in China. 

The Shidaowon site also has two Guohe One demonstration reactors. These are CAP1400 reactors, enlarged versions of the CAP1000 pressurized water nuclear reactors developed in collaboration with US-based Westinghouse.

The CAP1000 is one of the reactors China plans to deploy widely, while also exporting it to other countries interested in building nuclear energy assets. 

In addition to providing countries with easy ways to switch to renewable energy, China will now also help them switch to nuclear power without high input costs. 

🔗 Sumber: interestingengineering.com