📌 TOPINDIATOURS Eksklusif ai: MIT Researchers Unveil “SEAL”: A New Step Towards Se

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 Breaking ai: Europe’s biggest vanadium battery goes live in Spain

Spain has recently wrapped up operational testing of the biggest vanadium flow battery for applied research on the old continent, thus marking a turning point for sustainable, long-duration energy storage.

The testing, carried out at the technology center in Cubillos del Sil, northwestern Spain, was completed by the government-backed research institution Fundación Ciudad de la Energía (Ciuden).

Established in 2006, the foundation validated a 1 MW/8 MWh vanadium redox flow battery (VRFB) system, capable of delivering one megawatt of power and storing eight megawatt-hours of energy.

As per Cuiden, the installation is designed not just to store energy, but to serve as an experimental platform for advanced storage technologies. “This 1MW, 8MWh energy storage system includes a 100kW/800kWh experimental module that will allow for various R&D tests,” the foundation reported.

Inside the VRFB system

Vanadium redox flow batteries are long-duration, rechargeable energy storage systems that use vanadium ions in liquid electrolytes to store power in external tanks rather than solid electrodes.

Compared to traditional lithium-ion (Li-ion) technologies, the new system can deliver power for more than 15 hours.

This reportedly makes it the longest-duration battery currently available in Spain for experimental research. According to Cuiden, the project is part of a broader effort to build a hybrid energy testbed that brings together several technologies.

Apart from the vanadium system, the site hosts a one-megawatt, 5.8 megawatt-hour (1 MW/5.8 MWh) sodium-sulfur battery and a 600-kilowatt, 1.3 megawatt-hour (600 kW/1.3 MWh) lithium-ion system.

Paired with a 2.2 megawatt (MW) solar installation, the setup offers nearly 15 megawatt-hours (MWh) of storage capacity. It can store the plant’s full daily output during peak generation periods.

“The contract, worth EUR 6,4 million [USD 7.4 million] was awarded to the Spanish company CYMI and incorporates South Korean technology from H2 Inc,” Cuiden said in a press release.

Experimental storage hub

To store energy, the vanadium system uses liquid electrolytes with vanadium ions in different oxidation states. These liquids are stored in external tanks and allow the system’s energy capacity and power output to be scaled independently.

The design delivers greater durability than conventional systems. “This gives it a long lifespan of over 20 years and allows for power-energy decoupling, making it easy to increase storage capacity,” the foundation revealed.

The facility is also exploring how storage technologies interact with emerging hydrogen systems. For this purpose, the team has integrated two electrolyzers, including a 300-kilowatt proton exchange membrane unit and a 250-kilowatt solid oxide electrolyzer (SOEC), into the site.

The combination is expected to deliver a unique testing environment where solar power, batteries, as well as hydrogen production can be studied together.

The project is funded under the NextGenerationEU recovery program, which is a temporary pandemic recovery instrument made to rebuild a greener, digital, and more resilient European Union. It is part of Spain’s broader plan to modernize its energy infrastructure.

“[The initiative] aims to obtain technical data for the industrial-scale development of the various technologies that allow the extrapolation of their optimal operating conditions, thus promoting the decarbonization of industry,” Ciuden concluded.

🔗 Sumber: interestingengineering.com