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

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

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 Hot ai: Video: China’s six-foot humanoid robot knocks down sandbags wi

Another Chinese robotics player, Unitree, is pushing the limits of humanoid robotics.

On January 4, the company released a daily training video of its H2 robot, showcasing remarkable agility with moves like flying kicks, backflips, and sandbag strikes.

The footage showcases rapid advances in robot mobility and martial skills, amid competing firms also releasing videos claiming similar breakthroughs.

In a recent showcase of H2 uploaded in December, the 6-foot humanoid demonstrated its powerful actuation, performing dynamic moves such as punches, kicks, and knee strikes.

Robotics gets fierce

The showcase highlighted H2’s impressive agility as it performs flying kicks, executes backflips, and delivers powerful sandbag strikes, demonstrating advanced balance, coordination, and high-performance actuation in dynamic combat-style movements.

The YouTube video description read, “Have you exercised today? How about training together with a robot? Please use robots in a friendly and safe manner, and keep a safe distance.”

The H2 is Unitree Robotics’ tallest and most advanced humanoid to date, measuring nearly 6 feet (180 cm) and weighing about 154 pounds (70 kg). Visually, it marks a clear departure from earlier designs, featuring a stylized, silver, human-like face with defined eyes, lips, and a nose, giving it a more expressive, lifelike appearance.

The H2 follows the widely recognized H1 humanoid, which gained attention for its speed and agility during China’s 2024 Spring Festival Gala. While recent demonstrations have focused on the H2’s striking power and aggressive movements, the robot’s most significant innovations lie beneath the surface.

At its core, the H2 is driven by 31 degrees of freedom and exceptionally high joint output, delivering up to 360 Nm of torque. These capabilities are coordinated by advanced motion control algorithms that allow the robot to reproduce complex, dynamic actions with balance and precision accurately. This combination of hardware strength and software intelligence enables the fluid execution of demanding movements.

Dexterity meets control

Beyond its striking physical demonstrations, the H2 introduces a significant functional upgrade with newly designed, dexterous hands.

The humanoid’s arms now offer a full seven degrees of freedom, up from the four DOF seen in earlier models. This human-like articulation marks a shift in focus from basic locomotion toward complex manipulation, positioning the H2 as a true working robot rather than a simple moving platform.

With improved gripping and handling abilities, the humanoid is designed for practical roles in factories and logistics environments, with long-term potential for use in domestic settings.

Another notable detail emerging from recent footage is a glimpse into Unitree’s teleoperation system. While full autonomy remains the ultimate objective, remote control through human operators is increasingly viewed as a critical transitional technology. Teleoperation allows robots to perform useful tasks in real-world environments before fully autonomous systems are mature.

Unitree has been developing platforms that enable operators to control the robot in real time using wearable control rigs or mixed-reality devices such as the Apple Vision Pro. This approach allows human judgment and dexterity to be directly translated into robotic motion, improving task reliability and safety.

The broader humanoid robotics sector is advancing quickly. Recent demonstrations from other companies, including footage of Figure’s humanoid robot jogging at near-human speeds, highlight rapid progress in both locomotion and control.

Such videos show smooth acceleration, precise directional changes, and controlled braking in complex environments, underscoring how quickly humanoid capabilities are approaching human-like performance.

🔗 Sumber: interestingengineering.com