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

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

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: Video: South Korea’s KAIST humanoid robot dances and shoots

A field test video depicts South Korea’s KAIST Humanoid v0.7 robot outstanding moonwalk and soccer skills, thus highlighting its precise high-speed locomotion.

The 165-lb (75-kilogram) humanoid was developed at the Dynamic Robot Control & Design Laboratory (DRCD Lab), at the Korea Advanced Institute of Science and Technology (KAIST), under the leadership of Hae-Won Park, PhD.

The clip highlights the humanoid’s use of Physical AI, an approach which enables autonomous machines to perceive, interpret, and execute complex actions in real-world environments. It can be seen running, jumping, taking shots on goal, and performing fluid dance movements on a soccer pitch.

According to the research team, the five-foot-five-inch robot combines advanced hardware with intelligence control systems. It is designed to be reliable, efficient and scalable.

Physical AI in action

Instead of relying on off-the-shelf-parts, the Korean researchers independently developed all of v0.7’s components, including the motors, gearboxes and motor drivers, thus making it technically independent.

This made it possible to optimize torque density and power-to-weight ratios. As per the team, both are critical for high-speed locomotion and dynamic balance.

At the same time, the robot’s actuation system is based on a Quasi-Direct Drive (QDD) architecture. It pairs high-torque motors with low gear ratios, and boosts responsiveness while enabling more precise control.

This is supported by a custom-designed 3K compound planetary gearbox, which achieves high gear reduction in a compact, single-stage configuration. It results in a lighter, more efficient system capable of handling demanding tasks like running, jumping and rapid directional changes.

Hae-Won Park emphasized that the robot can run at speeds of up to 10.7 feet per second (about 7.3 miles per hour or 12 kilometers per hour) on flat ground. It can also climb steps over 12 inches (30 centimeters) high.

“The team plans to further enhance its performance, aiming for a driving speed of 4.0 m/s (approximately 14 km/h), ladder climbing, and over 40-centimeter step-climbing capability,” Park said.

What’s more, the humanoid’s knee actuator can deliver up to 320 Newton-meters (Nm) of peak torque. Meanwhile, the ankle actuator is optimized for fast response and stability. This allows the system to perform complex movements with both strength and finesse.

A powerful software

To produce smoother and more natural behavior, the researchers integrated deep reinforcement learning (DRL) with human motion data. They trained the system in simulation and used human movement as a behavioral prior. This is how v0.7 can avoid the jerky motions often seen in purely AI-driven systems.

It also incorporates Motor Operating Region (MOR) modeling, which constrains the simulation to match the physical limits of the hardware, and adopts a hybrid approach known as modular residual learning.

“This achievement is an important milestone that has achieved independence in both hardware and software aspects of humanoid research by securing core components and AI controllers with our own technology,” Park said.

The robot can also navigate uneven terrain using proprioception alone, without relying on visual sensors. This is particularly relevant for industrial environments where visibility may be limited.

“We will further develop it into a complete humanoid including an upper body to solve the complex demands of actual industrial sites and furthermore, foster it as a next-generation robot that can work alongside humans,” Park concluded in a press release.

The DRCD Lab is meanwhile developing DynaFlow, a framework aimed to enable robots to learn complex tasks directly from human demonstrations. It could allow humanoids to perform practical jobs, from handling tools to operating machinery.

🔗 Sumber: interestingengineering.com