📌 TOPINDIATOURS Update ai: MIT Researchers Unveil “SEAL”: A New Step Towards Self-

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

📌 TOPINDIATOURS Update ai: Japan: Scientists boost solar hydrogen output by captur

Researchers in Japan have developed a new photocatalyst that can use a much broader range of sunlight to produce hydrogen, offering a potential boost to clean fuel technologies.

The breakthrough comes from the Institute of Science Tokyo, where scientists designed a dye-sensitized system that captures long-wavelength visible light that conventional photocatalysts typically miss.

By rethinking the metal at the heart of the light-absorbing dye, the team has demonstrated a way to improve solar-to-hydrogen conversion without increasing the overall system’s complexity. The work points to a practical path for making hydrogen production from sunlight more efficient under real-world conditions.

Limits of conventional solar hydrogen systems

Solar hydrogen production relies on photocatalysts that absorb sunlight and use that energy to split water into hydrogen and oxygen. This approach is attractive because it can generate fuel without carbon emissions. However, many existing photocatalysts face a fundamental limitation. They only absorb a narrow portion of visible light.

Most traditional systems respond mainly to shorter visible wavelengths. As a result, a large share of incoming solar energy passes through unused. This inefficiency becomes more noticeable in low-light conditions or when it’s cloudy, where every photon counts.

Dye-sensitized photocatalysts were developed to address part of this problem. In these systems, a dye molecule absorbs light and transfers the energy to a catalyst surface. The dye acts like a light-harvesting antenna, improving performance compared to bare catalysts. Still, even these systems often rely on dyes with limited absorption ranges.

Switching the metal to expand light absorption

The Science Tokyo team, led by Professor Kazuhiko Maeda and graduate student Haruka Yamamoto, focused on the dye itself. Most dye-sensitized photocatalysts use ruthenium-based complexes, which absorb visible light only up to about 600 nanometers.

“Dye-sensitized photocatalysts typically use ruthenium complexes as the photosensitizing dyes. However, ruthenium-based complexes typically absorb only shorter visible wavelengths up to 600 nm,” explained Maeda.

To overcome this limit, the researchers replaced ruthenium with osmium at the center of the dye complex. This single change significantly altered the light absorption behavior. The new osmium-based dye can absorb visible light from 600 to 800 nanometers, covering a much larger portion of the solar spectrum.

This broader absorption means the system can generate more excited electrons when exposed to sunlight, directly improving hydrogen evolution.

Why does osmium make a difference?

The performance gain comes from a property known as the heavy-atom effect. Osmium promotes singlet–triplet excitation, a type of low-energy electronic transition that allows absorption of longer-wavelength light. Ruthenium complexes cannot access these transitions as effectively.

“In our efforts to extend the range of light absorption, osmium proved to be a key element in accessing wavelengths that ruthenium complexes could not use, leading to a two-fold increase in hydrogen production efficiency,” noted Maeda.

According to the study, the new system achieved up to twice the solar-to-hydrogen conversion efficiency of traditional dye-sensitized photocatalysts. Importantly, this improvement does not rely on intense sunlight, making it more suitable for practical outdoor use.

Implications for future clean energy technologies

The ability to convert more incoming photons into chemical energy could benefit a range of solar-driven technologies. Systems designed for artificial photosynthesis or decentralized hydrogen production could operate more reliably across different climates and lighting conditions.

While further optimization is still needed, the research establishes a clear design strategy for next-generation photocatalysts. By carefully selecting metal centers that extend light absorption, scientists can unlock more of the sun’s energy without redesigning entire systems.

The study was published in the journal ACS Catalysis.

🔗 Sumber: interestingengineering.com