📌 TOPINDIATOURS Eksklusif ai: Zuckerberg Already Blowing Up Relationship With New

This summer, Meta CEO Mark Zuckerberg moved heaven and earth to recharge his company’s lagging AI division, spending billions of dollars to poach top talent across the industry. But it sounds like he’s already falling out with the man he hired to spearhead Meta’s renewed AI efforts, Alexandr Wang.

According to new reporting from the Financial Times, things are tense between the duo. Sources familiar with the matter say that Wang has told associates he finds Zuckerberg’s micromanagement to be suffocating. Meanwhile, some staff wonder if the 28-year-old Wang is out of his depth, lacking both the expertise and experience to lead such a colossal effort.

It’s another sign of Zuckerberg’s direction for the company coming under the microscope, after his last big gamble — on creating a virtual reality “Metaverse” — failed spectacularly. Market skepticism was in full display in October, when Meta announced billions more in AI spending this year and the next, causing its stock to plunge 11 percent and erase over $200 billion in market cap.

His decision to hire Wang was controversial to begin with. Wang is the founder and former CEO of the AI data annotation startup Scale AI, which provides an essential service for training AI models, but doesn’t actually build them. In June, Zuckerberg poured over $14 billion into Scale AI, and poached Wang in the process to lead Meta’s newly dubbed Superintelligence Labs. Wang is also leading a secretive “TBD” (To Be Determined) lab which works in its own building.

The cost of the hire wasn’t just monetary: Meta’s then-chief AI scientist, Yann LeCun, didn’t take kindly to being forced to start reporting to Wang, and made a shock exit in November. LeCun is considered a “godfather” of the field for his pioneering work on neural networks, and he likely felt insulted to see his research rather than product-focused AI lab being hollowed out by firings while Zuckerberg offered astronomical nine-figure contracts to bring in talent to Wang’s Superintelligence Labs.

The new lab will double down on using large language models, the same architecture that powers AI chatbots like ChatGPT and Gemini, in its efforts to build a “superintelligence” that equals or surpasses human capabilities. LeCun viewed LLMs as a dead end and favored building new kinds of AI models instead.

Some staff questioned Wang’s credentials to lead such an important effort, and it’s a fair point to raise because Wang’s company didn’t build AI models at all. Instead, its focus was on the data used to train them, an entirely different kettle of fish. Now, he’s expected to not just develop AI models, but the kind that would rival human intelligence.

This isn’t the first sign of flair-ups with Wang at the company. The New York Times reported earlier this month that Wang was clashing with longtime Zuckerberg lieutenant Chris Cox, who wanted to focus on using Facebook and Instagram to train Meta’s new model. Wang disagreed and argued for focusing on catching up with Google and OpenAI’s models instead.

Zuckerberg’s determination is to build these products as fast as possible, which will only heighten tensions. “If you build too slowly,” he said in an interview on the Access podcast, per the FT, ”then you are just out of position on what I think is going to be the most important technology that enables the most new product and innovation and value creation in history.”

His other big new hire, former Github CEO Nat Friedman, is in a similar position to Wang, and some in his team were frustrated after what they perceived was the rushed release of Vibes, a feed of AI-generated videos akin to OpenAI’s Sora 2 app.

The pressure isn’t going to relent anytime soon. The secretive TBD lab is aiming to release an entirely new AI model built from scratch in the first quarter next year, people familiar with the matter told FT.

More on Meta: Meta’s $27 Billion Datacenter Is Wreaking Havoc on a Louisiana Town

The post Zuckerberg Already Blowing Up Relationship With New Head of AI He Paid Ten Zillion Dollars to Hire appeared first on Futurism.

🔗 Sumber: futurism.com

📌 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