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

📌 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 Update ai: Elon Musk’s Starship Explosion Endangered Hundreds of Airli

On January 16, SpaceX conceded that the latest prototype of its enormous Starship spacecraft had “experienced a rapid unscheduled disassembly during its ascent burn” — a tongue-in-cheek admission that the massive rocket had exploded mid-flight.

Countless videos circulating online showed a massive stream of reentering pieces of the Starship rocket blazing across the evening sky over the West Indian islands of Turks and Caicos. It was a dazzling sight to behold as the destruction streaked across the sky, like something from science fiction.

“Success is uncertain, but entertainment is guaranteed!” a gleeful Musk wrote at the time.

But it wasn’t all fun and games. Residents of Turks and Caicos, for instance, soon found scraps of burnt rubber and other rocket pieces littering the Caribbean islands’ otherwise pristine beaches.

The incident, which was almost perfectly repeated on February 24, even forced airlines to quickly adjust their flight paths to avoid the terrifying field of burning metal.

And now, eleven months later, the Wall Street Journal has obtained Federal Aviation Administration documents revealing that three aircraft — a JetBlue passenger plane, an Iberia Airlines flight, and a private jet, carrying a total of 450 people total — were in far greater danger than SpaceX and government officials let on at the time.

The reporting also highlights the possibility that SpaceX CEO Elon Musk’s extremely close relationship and ample influence in Washington, DC, may have played a role in FAA officials looking the other way as the company’s Starship rockets kept exploding during tests.

While all three planes landed safely on January 16, if any one of them had been struck by a piece of Starship debris, it could’ve been a major disaster.

Air traffic controllers had to scramble to ensure the planes were far away from the debris field, leading to an increase in their workload, a “potential extreme safety risk,” according to an FAA report obtained by the WSJ.

Air traffic controllers were left bewildered.

“I don’t know if you guys were advised, but there was a rocket launch and apparently the rocket exploded and there was debris in the area between us and Miami which basically covers our entire airspace,” one controller in the area said. “So I need to keep all the aircraft clear of that area because of the debris.”

SpaceX failed to inform the agency through its official hotline that its Starship had exploded, per the documents. The hotline is designed to alert officials at the FAA so they can act in time and keep pilots away from danger.

Previously determined no-fly debris zones were activated four minutes after SpaceX lost contact with its Starship vehicle. The Elon Musk-led firm didn’t confirm with the FAA that it had lost the craft until 15 minutes later.

SpaceX has since denied that anybody was ever in danger in an excoriating statement posted to Musk’s social media platform X.

“Yet another misleading ‘story’ the company’s official account wrote, echoing the voice of its mercurial CEO, who has an extremely troubled relationship with the news industry. “The reporters were clearly spoon-fed incomplete and misleading information from detractors with ulterior motives.”

“SpaceX is committed to responsibly using airspace during launches and reentries, prioritizing public safety to protect people on the ground, at sea, and in the air,” the statement reads.

The FAA suspended a safety review in August, even though its own policies call for addressing safety risks, per the WSJ, arguing that safety recommendations were already being implemented.

Meanwhile, SpaceX continues to make improvements to its Starship super heavy launch platform through an unusual iterative design approach that has already led to over a dozen massive explosions.

How future tests will fare — or if the public will ever be in danger — remains to be seen.

The company’s most recent, eleventh full-scale test took place on October 13. The rocket harmlessly splashed down in the Indian Ocean after successfully reaching orbit.

A new and even more powerful version is expected to launch sometime early next year.

More on Starship: Elon Says His New Rocket Is as Important as the Origin of Life Itself

The post Elon Musk’s Starship Explosion Endangered Hundreds of Airline Passengers appeared first on Futurism.

🔗 Sumber: futurism.com