📌 TOPINDIATOURS Eksklusif ai: MIT Researchers Unveil “SEAL”: A New Step Towards Se

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

📌 TOPINDIATOURS Breaking ai: It Turns Out Waymos Are Being Controlled by Workers i

Waymo has established itself as the autonomous ride-hailing service to beat, operating a fleet of several thousand self-driving taxis across the United States, with active services in ten major metropolitan areas.

That “self-driving” may be due for some extra scrutiny, though. During a Congressional hearing on Wednesday, Waymo’s chief safety officer, Mauricio Peña, was grilled over the company’s use of Chinese-made vehicles and reliance on overseas workers, as Business Insider reports.

The stakes and public safety implications are considerable. The news comes roughly a week after a Waymo robotaxi struck and injured a child near a Santa Monica, California, elementary school, triggering a federal probe.

After being pressed for a breakdown on where these overseas operators operate, Peña said he didn’t have those stats, explaining that some operators live in the US, but others live much further away, including in the Philippines.

“They provide guidance,” he argued. “They do not remotely drive the vehicles. Waymo asks for guidance in certain situations and gets an input, but the Waymo vehicle is always in charge of the dynamic driving tasks, so that is just one additional input.”

The admission didn’t sit well with senator Ed Markey (D-MA), who argued that “having people overseas influencing American vehicles is a safety issue.”

“The information the operators receive could be out of date. It could introduce tremendous cybersecurity vulnerabilities,” he argued. “We don’t know if these people have US driver’s licenses.”

“It’s one thing when a taxi is replaced by an Uber or a Lyft,” Markey concluded. “It’s another thing when the jobs just go completely overseas.”

Waymo has been fairly upfront about its human operators. In a May 2024 blog post, the company compared it to a “phone-a-friend.”

“When the Waymo vehicle encounters a particular situation on the road, the autonomous driver can reach out to a human fleet response agent for additional information to contextualize its environment,” the post reads. “The Waymo Driver [software] does not rely solely on the inputs it receives from the fleet response agent and it is in control of the vehicle at all times.”

In case the car’s driving software encounters something atypical, it may choose to send a request to a human fleet response agent, who then can help out by viewing real-time feeds from the vehicle’s exterior cameras.

While that may sound like the remote operator isn’t directly controlling the vehicle’s driving responses, it nonetheless goes to show how autonomous vehicles still rely substantially on human intellect. Fleet response agents may determine what lane a vehicle should pick, or propose a “path for the vehicle to consider,” as the blog post explains.

Put simply, the remote agent may not control the steering wheel, but they still make major decisions on where the vehicle navigates next.

During the same hearing, Tesla’s VP of vehicle engineering Lars Moravy told lawmakers that Tesla’s vehicles also rely on similar remote operators.

“We have many layers of security within our system and, similar to what Dr. Peña said, our driving controls, go, stop, steer, are in a core embedded central layer that cannot be accessed from outside the vehicle,” he said.

Moravy also argued that to stop anybody from taking control of vehicles, the company “actively participates in hacking events, trying paying people to try to get into our vehicles.”

The executives’ remarks are the latest illustration of how driverless taxis navigating public roads today are still far from being 100 percent autonomous. Tesla, in particular, has been playing it safe, quietly pausing its “unsupervised” robotaxi rides last week, meaning that there currently don’t appear to be any robotaxis with no human “safety monitor” in the driver’s seat.

It’s an especially glaring subject as lawmakers continue to ponder the risks of having autonomous vehicles coexist with human drivers on the road. And given the latest data, those risks remain substantial, with new National Highway Traffic Safety Administration data suggesting Tesla’s robotaxis are crashing three times as much as humans — even with human monitors.

Adding the influence of an entirely separate third party, a remote assistance operator who’s based overseas, could be a recipe for disaster, according to Markey, a glaring safety gap that needs to be filled.

“Overseas remote assistance operations may be more susceptible to physical takeover by hostile actors, potentially granting them driver-like control of thousands of vehicles transporting passengers on American roads,” he said in a statement. “Heavy and fast-moving vehicles could quickly become the weapons of foreign actors seeking to harm innocent Americans.”

More on Waymo: Waymo Under Investigation After Crashing Into Child Outside Elementary School

The post It Turns Out Waymos Are Being Controlled by Workers in the Philippines appeared first on Futurism.

🔗 Sumber: futurism.com