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 Breaking ai: Scientists Say Heck, Just Nuke a Killer Asteroid Heading

Plenty of asteroids can survive their fiery plunge through the Earth’s atmosphere. If they’re big enough, they can prove incredibly destructive, like the 60-foot Chelyabinsk meteor that exploded over the southern Ural region in Russia in 2013, releasing a blast equivalent to 30 times the energy of the atomic bomb that was dropped on Hiroshima.

And in case an even larger space rock were to ever threaten humanity, we’d have to get creative to keep it from colliding with our planet. Crashing a spacecraft into it like a pool ball to redirect its path — just like NASA did with its proof of concept Double Asteroid Redirection Test (DART) mission in 2022 — may not always be on the table, given the many uncertainties involved.

In a new paper published in the journal Nature Communications, an international team of researchers — including scientists from CERN and the University of Oxford — revisited the idea of blowing up an incoming asteroid with a nuclear warhead.

There are intuitive concerns. What if the asteroid shattered, turning a cosmic sniper shot into a shotgun blast of debris raining down over our planet?

But the team used CERN’s Super Proton Synchrotron (SPS) to study how asteroid materials react to different levels of physical stress, including large-scale simulations of nuclear deflection, and found that the space rocks are surprisingly resilient.

“Planetary defense represents a scientific challenge,” said Karl-Georg Schlesinger, cofounder of nuclear deflection startup Outer Solar System Company (OuSoCo), which partnered with the scientists, in a statement. “The world must be able to execute a nuclear deflection mission with high confidence, yet cannot conduct a real-world test in advance.”

In an experiment, the team exposed samples of a metal-rich meteorite to 27 short but intense pulses of a proton beam at CERN’s HiRadMat facility. Afterward, the team moved the meteorite to the ISIS Neutron and Muon Source at the Rutherford Appleton Laboratory in the UK to analyze changes to its internal structure at a microscopic level.

To their surprise, the “material became stronger, exhibiting an increase in yield strength, and displayed a self-stabilizing damping behavior,” explained OuSoCo cofounder Melanie Bochmann.

The finding could have major implications for how we approach future asteroid redirection efforts.

“Our experiments indicate that — at least for metal-rich asteroid material — a larger device than previously thought can be used without catastrophically breaking the asteroid,” Bochmann said. “This keeps open an emergency option for situations involving very large objects or very short warning times, where non-nuclear methods are insufficient and where current models might assume fragmentation would limit the usable device size.”

Fortunately, the researchers could soon have far more data to go by. Both NASA and the European Space Agency are planning to study Apophis, an enormous asteroid somewhere between 1,000 and 1,500 feet in width, which is expected to come eerily close to the Earth — closer than many geosynchronous satellites at just 20,000 miles — to Earth in April 2029.

“As a next step, we plan to study more complex and rocky asteroid materials,” the researchers said in a statement. “One example is a class of meteorites called pallasites, which consist of a metal matrix similar to the meteorite material we have already studied, with up to centimeter-sized magnesium-rich crystals embedded inside.”

The upcoming research could have fascinating implications outside of asteroid redirection as well.

“Because these objects are thought to originate from the core–mantle boundary of early planetesimals,” they added, “such experiments could also provide valuable insights into planetary formation processes.”

More on asteroids: Asteroid Behaving Strangely

The post Scientists Say Heck, Just Nuke a Killer Asteroid Heading for Earth appeared first on Futurism.

🔗 Sumber: futurism.com