📌 TOPINDIATOURS Hot ai: MIT Researchers Unveil “SEAL”: A New Step Towards Self-Imp

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: World’s longest expressway tunnel opens in China at a

The Tianshan Shengli Tunnel, a 13.75-mile (22 km) long tunnel at the center of the Urumqi-Yuli Expressway, and the world’s longest, is now open in China.

Constructed over five years using innovative approaches, the tunnel halves the travel time between the regional capital, Urumqi, and the city of Korla to 3.5 hours, state media reported. 

In the past few decades, China has undertaken some of the largest infrastructure projects in the world. From the 22.5 gigawatt hydroelectric power plant at the Three Gorges Dam to the Pinglu Canal, which provides an inland-to-sea connection for large ships, China is making huge leaps in the ambition of its projects. 

Interesting Engineering has previously reported how the Asian giant is also using its deserts to field large-scale solar power plants, and the Belt and Road Initiative (BRI) looks to connect the nation with Africa and Europe through a network of new railways, ports, and roads. 

The Tianshan Shengli Tunnel might seem like a minor project compared to these mega projects. However, it is still a significant achievement for the local region and set world records during construction. 

Where is the tunnel? 

The tunnel is located in the Xinjiang Uyghur Autonomous Region and facilitates a drive through the Tianshan Mountains. The region shares borders with eight countries, including Kazakhstan, Kyrgyzstan, Tajikistan, and Pakistan, facilitating connections to Central Asia. 

Before the expressway, travelling between Urumqi and Koral would take nearly seven hours. But now the travelling time will be reduced to 3.5 hours.

Additionally, it connects the region to various economic corridors across the country, aligning it with the national ‘dual circulation’ strategy, which seeks to integrate domestic and foreign trade. 

Work on the tunnel began in April 2020, but engineers faced challenges with terrain and environmental conditions during construction. 

Innovations in its construction

The tunnel runs through the Tianshan Mountains at an altitude of 9,842 feet (~ 3,000 meters) above sea level. Temperatures at these altitudes reach a bone-chilling minus 43.6°F ( minus 42 °C).

Attempting to construct an expressway using conventional methods would have taken Chinese engineers at least a decade to complete, the chief engineer of the Xinjiang Transport Investment and Development arm told local media. 

So, the engineers used a novel “three tunnels plus four shafts” strategy during the construction. Instead of trying to build a long, deep main tunnel, the engineers dug three tunnels: the main one and two parallel ones. 

The parallel tunnels facilitated geological investigation ahead of the main bore and provided workers with access to the site and equipment. In emergencies, the third tunnel could be used to house ventilation systems and serve as an emergency escape route. 

The four shafts were vertical passages dug from the surface to the tunnel depth. Shafts nearly 2,300 feet (700 m) deep were dug and served as additional entry and exit points, allowing work to proceed in parallel rather than at just two ends.

In addition to being the world’s longest expressway tunnel, the construction also holds the world record for the world’s deepest vertical shaft for a highway tunnel.  

Reduced travel between the two regions will facilitate the flow of energy and manufactured goods to the north and agricultural goods to the south, the South China Morning Post reported. 

🔗 Sumber: interestingengineering.com