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

📌 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 Breaking ai: Europe’s most powerful rocket to lift off with four boost

Europe will launch its most powerful rocket to date, the four-booster Ariane 64, from French Guiana today.

It will be the first time the Ariane 6 rocket launches in this configuration, as well as the first time it flies a commercial customer payload to space.

The rocket is scheduled to launch 32 Amazon Leo satellites into orbit. Amazon has 18 Ariane 6 launches booked to support the deployment of its Starlink-rivaling constellation.

Ariane 64: Europe’s most powerful rocket

The mission, called VA267 (LE-01 for Amazon Leo), will see Ariane 6 fly in its full-power Ariane 64 configuration for the very first time. Using four boosters, the rocket is capable of carrying more than 20 metric tons to orbit. It will also be roughly twice as powerful as the two-booster version that has flown before.

This makes it Europe’s most powerful rocket ever, though it still lags behind SpaceX’s most powerful rockets. As a point of reference, Ariane 64 will produce an impressive 3.5 million lbf of thrust at launch. SpaceX’s Falcon Heavy produces approximately 5.1 million lbf, while Starship’s 33 Raptor engines produce a staggering 16.5 million lbf.

“It’s a special launch—something new for us on Ariane 6,” ArianeGroup Chief Technical Officer Hervé Gilibert said during a press event earlier this month, according to EuroNews.

“Don’t be surprised if you see it accelerate much more than Ariane 62, the version we have already launched five times,” he continued. “It delivers significantly more power, allowing much heavier payloads to be sent into space.”

The VA267 mission will see 32 Amazon Leo satellites stored under a 20-meter fairing. If all goes to plan, they will be deployed by the rocket’s upper stage in low Earth orbit. To date, Amazon has launched more than 150 of its Leo, formerly Project Kuiper, satellites to orbit. The delivery giant aims to eventually fly over 3,200 of the satellites to space. VA267 will be the first launch for the constellation performed by a European rocket.

According to a statement from Arianespace, the mission will last a total of 1 hour and 54 minutes, from launch to deployment of all the satellites. During that time, it will perform nearly a full orbit around our planet.

Watch Ariane 6 fly skyward

The Ariane 6 rocket, as well as its engines and avionics, is built across 13 different nations, all of which are members of the European Space Agency. “We are working with more than 600 subcontractors,” Gilibert explained during the press event.

The rocket was designed to halve the operating costs of its predecessor, Ariane 5. However, the rocket is still fully expendable, unlike SpaceX’s partially reusable Falcon 9 launch vehicles and its fully reusable Starship.

Ariane 6 is roughly 62 metres (203 feet) tall, which is approximately the height of a 20-story building. To lift itself to space, the rocket will consume 142,000 kilograms (313,056 pounds) of solid propellant before burning out roughly two minutes after launch. You can watch the launch live via the Arianespace video below.

🔗 Sumber: interestingengineering.com