📌 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: Replacing coders with AI? Why Bill Gates, Sam Altman

In the race to automate everything – from customer service to code – AI is being heralded as a silver bullet. The narrative is seductive: AI tools that can write entire applications, streamline engineering teams and reduce the need for expensive human developers, along with hundreds of other jobs. 

But from my point of view as a technologist who spends every day inside real companies’ data and workflows, the hype doesn’t match up with the reality. 

I’ve worked with industry leaders like General Electric, The Walt Disney Company and Harvard Medical School to optimize their data and AI infrastructure, and here’s what I’ve learned: Replacing humans with AI in most jobs is still just an idea on the horizon. 

I worry that we're thinking too far ahead. In the past two years, more than a quarter of programming jobs have vanished. Mark Zuckerberg announced he is planning to replace many of Meta’s coders with AI. 

But, intriguingly, both Bill Gates and Sam Altman have publicly warned against replacing coders. 

Right now, we shouldn’t count on AI tools to successfully replace jobs in tech or business. That’s because what AI knows is inherently limited by what it has seen – and most of what it has seen in the tech world is boilerplate.

Generative AI models are trained on large datasets, which typically fall into two main categories: publicly available data (from the open internet), or proprietary or licensed data (created in-house by the organization, or purchased from third parties). 

Simple tasks, like building a basic website or configuring a template app, are easy wins for generative models. But when it comes to writing the sophisticated, proprietary infrastructure code that powers companies like Google or Stripe, there’s a problem: That code doesn’t exist in public repositories. It’s locked away inside the walls of corporations, inaccessible to training data and often written by engineers with decades of experience.

Right now, AI can’t reason on its own yet. And it doesn’t have instincts. It’s just mimicking patterns. A friend of mine in the tech world once described large language models (LLMs) as a "really good guesser." 

Think of AI today as a junior team member — helpful for a first draft or simple projects. But like any junior, it requires oversight. In programming, for example, while I’ve found a 5X improvement for simple coding, I’ve found that reviewing and correcting more complicated AI-produced code often takes more time and energy than writing the code myself. 

You still need senior professionals with deep experience to find the flaws, and to understand the nuances of how those flaws might pose a risk six months from now. 

That’s not to say AI shouldn’t have a place in the workplace. But the dream of replacing entire teams of programmers or accountants or marketers with one human and a host of AI tools is far premature. We still need senior-level people in these jobs, and we need to train people in junior-level jobs to be technically capable enough to assume the more complex roles one day. 

The goal of AI in tech and business shouldn’t be about removing humans from the loop. I’m not saying this because I’m scared AI will take my job. I’m saying it because I’ve seen how dangerous trusting AI too much at this stage can be. 

Business leaders, no matter what industry they’re in, should be aware: While AI promises cost savings and smaller teams, these efficiency gains could backfire. You might trust AI to perform more junior levels of work, but not to complete more sophisticated projects. 

AI is fast. Humans are smart. There’s a big difference. The sooner we shift the conversation from replacing humans to reinforcing them, the more we’ll reap the benefits of AI. 

Derek Chang is founding partner of Stratus Data.

🔗 Sumber: venturebeat.com