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

📌 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 Hot ai: Waymo CEO Says Society Is Ready for One of Its Cars to Kill So

Waymo robotaxis are so safe that, according to the company’s data, its driverless vehicles are involved in 91 percent fewer crashes compared to human-operated vehicles.

And yet the the company is bracing for the first time when a Waymo does kill somebody — a moment its CEO says society will accept, in exchange for access to its relatively safer driverless cars.

“We really worry as a company about those days,” said Waymo co-CEO Tekedra Mawakana on Monday during TechCrunch‘s Disrupt summit, as reported by SFGate. “You know, we don’t say ‘whether.’ We say ‘when.’ And we plan for them.”

“I think that society will,” she said in answer to a question on whether the public is prepared for a Waymo to cause a death. “I think the challenge for us is making sure that society has a high enough bar on safety that companies are held to.”

It’s strikingly mask-off utilitarian discourse from a tech CEO, but on those terms it’s also hard to argue with the logic: if Waymo’s cars are really safer than the average human driver, their widespread use would be a net good for traffic safety even if they still cause a trickle of accidents.

Driverless vehicles such as Waymo, started in 2009, are relatively new entrants to our roadways; lawmakers and companies are still rewriting the rules of the road when it comes to autonomous cars — so consider the society question unsettled, especially since the industry’s overall record is quite spotty.

For example, Waymo’s rival Tesla has notched at least three crashes involving its robotaxi service, not to mention lawsuits the company had to settle this year after Tesla’s Autopilot was accused of being at fault in two fatal incidents.

General Motors’ robotaxi competitor Cruise recently restarted operations after a pause following a 2023 incident when a Cruise robotaxi dragged a woman down a San Francisco street for 20 feet; that woman settled for millions after she sued the company.

But if anybody can handle any safety issues, Waymo seems heads and tails above its competition because its rollout into city streets has been extremely slow and deliberate; Mawakana said at the TechCrunch summit that the company continually retests its vehicles in order to address challenges that pop-up such as Waymos accidentally blocking emergency vehicles.

“We need to make sure that the performance is backing what we’re saying we’re doing,” she said.

(Complicating matters: self-driving car companies including Waymo still employ “remote operators” who can control a car in trouble from afar if it runs into trouble; it’s unclear how often these unseen employees need to step in, and whether reducing their role as robotaxis go mainstream would affect safety statistics.)

And even with its above-average safety record, Waymos have been known to behave in inexplicable ways, such as when one passed a stopped school bus that was unloading kids in Atlanta. That’s a violation that normally garners $1,000 fine and a court hearing, but nothing was issued to the company.

“These cars don’t have a driver, so we’re really going to have to rethink who’s responsible,” said Georgia state Representative Clint Crowe to Atlanta news station, KGW8.

More on Waymo: Family Baffled By Waymo Robotaxis Constantly Hanging Out in Front of Their House

The post Waymo CEO Says Society Is Ready for One of Its Cars to Kill Someone appeared first on Futurism.

🔗 Sumber: futurism.com