Join the event trusted by enterprise leaders for nearly two decades. VB Transform brings together the people building real enterprise AI strategy.\u00a0Learn more<\/em><\/p>\n\n\n\n Researchers at MIT have developed a framework called Self-Adapting Language Models (SEAL) that enables large language models (LLMs) to continuously learn and adapt by updating their own internal parameters. SEAL teaches an LLM to generate its own training data and update instructions, allowing it to permanently absorb new knowledge and learn new tasks.<\/p>\n\n\n\n This framework could be useful for enterprise applications, particularly for AI agents that operate in dynamic environments, where they must constantly process new information and adapt their behavior.<\/p>\n\n\n\n While large language models have shown remarkable abilities, adapting them to specific tasks, integrating new information, or mastering novel reasoning skills remains a significant hurdle.<\/p>\n\n\n\n Currently, when faced with a new task, LLMs typically learn from data \u201cas-is\u201d through methods like finetuning or in-context learning. However, the provided data is not always in an optimal format for the model to learn efficiently. Existing approaches don\u2019t allow the model to develop its own strategies for best transforming and learning from new information.<\/p>\n\n\n\n \u201cMany enterprise use cases demand more than just factual recall\u2014they require deeper, persistent adaptation,\u201d Jyo Pari, PhD student at MIT and co-author of the paper, told VentureBeat. \u201cFor example, a coding assistant might need to internalize a company\u2019s specific software framework, or a customer-facing model might need to learn a user\u2019s unique behavior or preferences over time.\u201d\u00a0<\/p>\n\n\n\n In such cases, temporary retrieval falls short, and the knowledge needs to be \u201cbaked into\u201d the model\u2019s weights so that it influences all future responses.\u00a0<\/p>\n\n\n\n \u201cAs a step towards scalable and efficient adaptation of language models, we propose equipping LLMs with the ability to generate their own training data and finetuning directives for using such data,\u201d the MIT researchers state in their paper.<\/p>\n\n\n\n The researchers\u2019 solution is SEAL, short for Self-Adapting Language Models. It uses a reinforcement learning (RL) algorithm to train an LLM to generate \u201cself-edits\u201d\u2014natural-language instructions that specify how the model should update its own weights. These self-edits can restructure new information, create synthetic training examples, or even define the technical parameters for the learning process itself.<\/p>\n\n\n\n Intuitively, SEAL teaches a model how to create its own personalized study guide. Instead of just reading a new document (the raw data), the model learns to rewrite and reformat that information into a style it can more easily absorb and internalize. This process brings together several key areas of AI research, including synthetic data generation, reinforcement learning and test-time training (TTT).<\/p>\n\n\n\n The framework operates on a two-loop system. In an \u201cinner loop,\u201d the model uses a self-edit to perform a small, temporary update to its weights. In an \u201couter loop,\u201d the system evaluates whether that update improved the model\u2019s performance on a target task. If it did, the model receives a positive reward, reinforcing its ability to generate that kind of effective self-edit in the future. Over time, the LLM becomes an expert at teaching itself.<\/p>\n\n\n\n In their study, the researchers used a single model for the entire SEAL framework. However, they also note that this process can be decoupled into a \u201cteacher-student\u201d model. A specialized teacher model could be trained to generate effective self-edits for a separate student model, which would then be updated. This approach could allow for more specialized and efficient adaptation pipelines in enterprise settings.<\/p>\n\n\n\n The researchers tested SEAL in two key domains: knowledge incorporation (the ability to permanently integrate new facts) and few-shot learning (the ability to generalize from a handful of examples).<\/p>\n\n\n\n For knowledge incorporation, the goal was to see if the model could answer questions about a text passage without having access to the passage during questioning. Finetuning Llama-3.2-1B on the raw text provided only a marginal improvement over the base model.\u00a0<\/p>\n\n\n\n However, when the SEAL model created \u201cself-edits\u201d by generating several \u201cimplications\u201d from a passage and was trained on this synthetic data, its accuracy jumped to 47%. Notably, this outperformed results from using synthetic data generated by the much larger GPT-4.1, suggesting the model learned to create superior training material for itself.<\/p>\n\n\n\n For few-shot learning, the researchers tested SEAL on examples from the Abstract Reasoning Corpus (ARC), where the model must solve visual puzzles. In the self-edit phase, the model had to generate the entire adaptation strategy, including which data augmentations and tools to use and what learning rate to apply.\u00a0<\/p>\n\n\n\n SEAL achieved a 72.5% success rate, a dramatic improvement over the 20% rate achieved without RL training and the 0% rate of standard in-context learning.<\/p>\n\n\n Some experts project that the supply of high-quality, human-generated training data could be exhausted in the coming years. Progress may soon depend on \u201ca model\u2019s capacity to generate its own high-utility training signal,\u201d as the researchers put it. They add, \u201cA natural next step is to meta-train a dedicated SEAL synthetic-data generator model that produces fresh pretraining corpora, allowing future models to scale and achieve greater data efficiency without relying on additional human text.\u201d<\/p>\n\n\n\n For example, the researchers propose that an LLM could ingest complex documents like academic papers or financial reports and autonomously generate thousands of explanations and implications to deepen its understanding.\u00a0<\/p>\n\n\n\n \u201cThis iterative loop of self-expression and self-refinement could allow models to keep improving on rare or underrepresented topics even in the absence of additional external supervision,\u201d the researchers explain.<\/p>\n\n\n\n This capability is especially promising for building AI agents. Agentic systems must incrementally acquire and retain knowledge as they interact with their environment. SEAL provides a mechanism for this. After an interaction, an agent could synthesize a self-edit to trigger a weight update, allowing it to internalize the lessons learned. This enables the agent to evolve over time, improve its performance based on experience, and reduce its reliance on static programming or repeated human guidance.<\/p>\n\n\n\n \u201cSEAL demonstrates that large language models need not remain static after pretraining,\u201d the researchers write. \u201cBy learning to generate their own synthetic self-edit data and to apply it through lightweight weight updates, they can autonomously incorporate new knowledge and adapt to novel tasks.\u201d<\/p>\n\n\n\n That said, SEAL is not a universal solution. For example, it can suffer from \u201ccatastrophic forgetting,\u201d where constant retraining cycles can result in the model learning its earlier knowledge.<\/p>\n\n\n\n \u201cIn our current implementation, we encourage a hybrid approach,\u201d Pari said. \u201cEnterprises should be selective about what knowledge is important enough to integrate permanently.\u201d\u00a0<\/p>\n\n\n\n Factual and evolving data can remain in external memory through RAG, while long-lasting, behavior-shaping knowledge is better suited for weight-level updates via SEAL.\u00a0<\/p>\n\n\n\n \u201cThis kind of hybrid memory strategy ensures the right information is persistent without overwhelming the model or introducing unnecessary forgetting,\u201d he said.<\/p>\n\n\n\n It is also worth noting that SEAL takes a non-trivial amount of time to tune the self-edit examples and train the model. This makes continuous, real-time editing infeasible in most production settings.<\/p>\n\n\n\n \u201cWe envision a more practical deployment model where the system collects data over a period\u2014say, a few hours or a day\u2014and then performs targeted self-edits during scheduled update intervals,\u201d Pari said. \u201cThis approach allows enterprises to control the cost of adaptation while still benefiting from SEAL\u2019s ability to internalize new knowledge.\u201d<\/p>\n Daily insights on business use cases with VB Daily<\/strong><\/p>\n If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI.<\/p>\n \n\t\t\t\t\tThanks for subscribing. Check out more VB newsletters here.\n\t\t\t\t<\/p>\n An error occured.<\/p>\n<\/p><\/div>\n
\n<\/div>The challenge of adapting LLMs<\/h2>\n\n\n\n
Creating self-adapting language models<\/h2>\n\n\n\n
![\"Overview](\"https:\/\/venturebeat.com\/wp-content\/uploads\/2025\/06\/image_b1ee4c.png?w=800\")
SEAL in action<\/h2>\n\n\n\n
![\"SEAL](\"https:\/\/venturebeat.com\/wp-content\/uploads\/2025\/06\/image_d93126.png?w=800\")
![\"SEAL](\"https:\/\/venturebeat.com\/wp-content\/uploads\/2025\/06\/image_1e87a7.png?w=800\")
![\"SEAL](\"https:\/\/venturebeat.com\/wp-content\/uploads\/2025\/06\/image_969ef3.png\")
Implications for the enterprise<\/h2>\n\n\n\n
Limitations of SEAL<\/h2>\n\n\n\n
![\"\"\/](\"https:\/\/venturebeat.com\/wp-content\/themes\/vb-news\/brand\/img\/vb-daily-phone.png\")
\n\t\t\t\t<\/div>\n<\/p><\/div>\n<\/div>\t\t\t<\/div>\r\n