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

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

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 Eksklusif ai: Interlocking materials offer high strength and flexibili

Researchers at the University of Colorado Boulder are developing a new class of “entangled materials” inspired by the surprising strength of a tangled ball of office staples. 

​​Much like a bird’s nest or a burr, a cluster of staples gains its strength from geometric interlocking rather than chemical bonds. But it retains the ability to instantaneously transition back into a loose piece through targeted vibration.

“We’ve been playing around with the idea of building blocks and geometry for many years, but we started looking at interlocking, entangled particles only recently,” said Professor Francois Barthelat, the leader of the Laboratory for Advanced Materials and Bioinspiration.

“We are excited about the combination of properties we can get out of these systems, and we believe this technology has the potential to go in many directions,” Barthelat added.

Geometry of grip

The work centers around “entanglement.” Through this, researchers are mimicking natural structures, such as bird nests and bone minerals, to create ultra-strong manufactured materials. 

Particle shape is key in this. As compared to smooth grains of sand that slide apart, specialized geometries allow individual pieces to physically intertwine. 

This mechanical locking creates a cohesive link that provides structural integrity without the need for adhesives.

“Let’s take sand as an example. Sand is smooth and convex-shaped, meaning it cannot interlock from grain to grain,” Youhan Sohn, Ph.D. student, said.

“However, we found that if we change the shape of a grain of sand, we can drastically affect its behavior and mechanical properties, including the particle’s ability to link with other particles,” Sohn explained.

For the study, Monte Carlo simulations were used to analyze particle geometry. and identified that “two-legged” staple shapes provide the most effective mechanical interlocking. 

Instead of stacking loosely, these U-shaped particles hook and weave into a singular mass that stubbornly resists being pulled apart.

Physical testing revealed that these entangled particles possess a rare dual advantage, maintaining simultaneous tensile strength and exceptional toughness.

Use of vibration

The real power of this material is in its response to a simple buzz.

Standard materials are permanent. For instance, a concrete bridge is there forever until it is smashed into dust. But Barthelat’s entangled particles are different.

The material’s standout feature is its capacity for rapid, reversible assembly controlled by vibrational patterns. 

Interestingly, the entanglement levels can be modulated on demand through these vibrations. Gentle frequencies can lock particles into a rigid structure, whereas more intense vibrations trigger the complete unraveling of the mass.

“It’s a strange material because it’s obviously not a liquid. However, it’s also not quite solid. This opens new and intriguing engineering possibilities,” Barthelat said. “Handling a bundle of these entangled particles feels very remote and exotic.”

Entangled materials offer potential for sustainability and advanced technology, particularly in civil engineering and robotics. 

It could enable large-scale structures, such as bridges, to be “unzipped” and recycled rather than demolished. Eventually, this technology could support a circular economy. 

Furthermore, it could advance swarm robotics, allowing fleets of small machines to interlock into functional tools and later disentangle to navigate tight spaces — a real-world parallel to the shape-shifting capabilities of cinematic sci-fi.

“Yes, kind of like that liquid metal T-1000 in Terminator 2, who can change shape to slide under a door and then transform back to a human’s size on the other side,” added Barthelat.

The researchers are currently pushing the boundaries of their work by testing multi-legged particle shapes modeled after high-grip plant burrs to achieve even more powerful entanglement. 

The study was published in the Journal of Applied Physics.

🔗 Sumber: interestingengineering.com