Posted on Leave a comment

A Survey on Self-Evolution of Large Language Models

Abstract

The self-evolution of large language models refers to the process where models iteratively improve their capabilities using self-generated data or feedback without relying on external human-annotated data. This paradigm has the potential to break through the bottleneck of human supervision, enabling LLMs to continuously and autonomously advance toward higher levels of intelligence. This paper systematically reviews the core concepts, mainstream technical approaches (including self-training, self-rewarding, self-reflection, and curriculum learning), representative works with their strengths and limitations, and finally discusses key challenges and future directions.

1. Introduction

As the capabilities of large language models such as GPT-4, Claude, and Llama rapidly increase, how to push models beyond the boundaries of existing training data has become a research focus. Traditional fine-tuning heavily relies on high-quality human-annotated data, which is costly and difficult to scale. Self-evolution emerges as a response—models leverage their own generation, reasoning, and evaluation abilities to build a closed-loop learning cycle, achieving “learning to become stronger.” Inspired by AlphaGo Zero’s self-play, this idea is rapidly developing in the natural language domain.

2. Core Framework of Self-Evolution

Self-evolution typically consists of four basic stages:

  • Data Generation: The model generates training samples (e.g., question-answer pairs, chain-of-thought reasoning paths) based on prompts or seed tasks.
  • Quality Filtering: Low-quality data is filtered out via rules, model self-scoring, or external validators.
  • Training Update: The filtered data is used for supervised fine-tuning (SFT) or reinforcement learning (RLHF/DPO).
  • Iterative Evaluation: The above process is repeated in subsequent rounds, forming an upward spiral.

3. Main Technical Approaches

3.1 Self-Training

The most classic method. The model first generates pseudo-labels from a small set of seed data, then selects high-confidence samples to add to the training set, and retrains. Representative work includes STaR (Self-Taught Reasoner): the model attempts to reason through problems; if it obtains the correct answer, the reasoning chain is retained as training data. Repeated iterations significantly improve mathematical reasoning ability. Limitations include susceptibility to amplifying initial biases and sensitivity to noise.

3.2 Self-Rewarding

To address the reliance on human-annotated reward models in RLHF, researchers propose letting the LLM act as both “actor” and “critic.” For example, Self-Rewarding LM: at each iteration, the model generates answers and scores its own responses according to predefined criteria, then uses these scores as reward signals for DPO training. Experiments on AlpacaEval show that alignment capability improves steadily over multiple rounds of self-rewarding iteration. However, self-scoring may suffer from preference collapse.

3.3 Self-Reflection

The model reviews its erroneous outputs and produces improved versions. A typical example is Reflexion: after failing a task, the agent converts error information and environmental feedback into textual memory, references these reflections in subsequent attempts, and gradually increases success rates. This approach performs well in interactive tasks such as programming and decision-making. A limitation is that reflection quality depends on the model’s original understanding ability.

3.4 Curriculum Learning and Progressive Evolution

Inspired by human learning, models start self-evolution from simple tasks and gradually progress to complex ones. For example, Self-Play Fine-Tuning (SPIN): the model plays against its older version, generating increasingly difficult adversarial examples, forcing itself to learn finer discrimination capabilities. This method yields significant gains in mathematical reasoning and instruction following.

4. Overview of Representative Works

MethodCore MechanismApplicable ScenarioKey Results
STaRSelf-generated reasoning chains + answer verificationMathematical reasoningSignificant accuracy improvement on GSM8K
Self-Rewarding LMSelf-scoring + DPOAlignment optimizationIterative growth on AlpacaEval scores
ReflexionError reflection + memory replayAgent tasksImproved success rates in coding, QA
SPINSelf-play + adversarial trainingGeneral capabilitiesContinuous score increase on MT-Bench

Additionally, OpenAI’s o1 model internally employs some form of self-search and reflection mechanism, further confirming the potential of the self-evolution approach.

5. Key Challenges

Despite promising prospects, self-evolution currently faces the following difficulties:

  • Mode Collapse: The model may fall into homogeneity during iterations; generated data lacks diversity, causing performance stagnation or even degradation.
  • Reward Hacking: Under self-rewarding mechanisms, the model learns to exploit scoring loopholes rather than genuinely improving capabilities.
  • Computational Cost: Each iteration requires full-scale training, resulting in huge resource consumption.
  • Evaluation Dilemma: Lack of independent, uncontaminated external benchmarks to measure real progress, as test sets may also be covered by model-generated data.

6. Future Directions

  • Hybrid Evolution: Combining limited human feedback with large-scale self-evolution to balance efficiency and stability.
  • Multimodal Self-Evolution: Incorporating visual, audio, and other signals into the closed loop to achieve cross-modal capability growth.
  • Open-Source Ecosystem Synergy: Community-driven self-evolution platforms allowing different models to distill and compete with each other.
  • Theoretical Understanding: Establishing convergence theories for self-evolution to guide algorithm design.

7. Conclusion

Self-evolution of LLMs is moving from concept to practice, demonstrating potential beyond static training in areas such as mathematical reasoning, code generation, and dialogue alignment. Although we are still far from fully “self-aware” evolution, this paradigm offers a feasible technical path to break through the data ceiling. Future research needs to find better trade-offs among stability, efficiency, and generalization to truly realize continuous autonomous growth of large models.


Note: This article is based on publicly available research up to 2025. For the latest advances, please refer to the most recent official publications.

Posted on Leave a comment

to edit with ai

Posted on Leave a comment

An AI video scriptwriting tool is in its early stages

Summary: An AI video scriptwriting tool is in its early stages: it is expected to be based on Douyin’s Jianying (CapCut) and WordPress, with the development language likely being Java.

Product Name: ScriptForge AI (or any placeholder name)

Description:

ScriptForge AI is a next-generation video scriptwriting assistant designed specifically for short-form video creators on platforms like Douyin (TikTok). It seamlessly integrates with Douyin’s Jianying (CapCut)​ for video editing and rendering, and uses WordPress​ as its content management and publishing backbone.

Built primarily with Java​ on the backend, the tool leverages natural language processing and large language models to help users generate compelling scripts, storyboards, shot lists, and dialogue automatically. Users can input a theme, genre, or key plot points, and the system outputs structured video scripts optimized for engagement metrics such as hook retention and pacing.

Key features include:

  • AI Script Generation: Produce 30–60 second short drama scripts with hooks, conflict escalation, and cliffhangers.
  • Shot-by-Shot Storyboard Export: Each scene includes camera angle, character action, lighting notes, and duration — ready to feed into CapCut.
  • WordPress Plugin Integration: Manage projects, store drafts, collaborate with team members, and publish final scripts directly as WordPress posts or pages.
  • CapCut Template Linking: Automatically map script segments to CapCut video templates for faster production.
  • Character Consistency Manager: Generate and store character reference images (via external AI image generators) and ensure visual continuity across scenes.
  • Multi-language Support: Generate scripts in Chinese, English, and other languages with localized cultural references.

Technical Stack:

  • Backend: Java (Spring Boot), RESTful APIs
  • Frontend: React-based dashboard embedded in WordPress admin panel
  • Database: MySQL / MariaDB (via WordPress)
  • AI Models: Integrated via cloud LLM APIs (e.g., DeepSeek, GPT)
  • Video Engine: CapCut SDK / API for template rendering

Target Audience: Independent short drama creators, social media content teams, and small studios looking to accelerate their pre-production workflow.

Monetization Model: Subscription-based (monthly/yearly) with tiered access to advanced features such as bulk generation, custom branding, and priority API calls.

Posted on Leave a comment

OpenAI is secretly advancing the development of its own web search service

Earlier reports revealed that OpenAI is secretly advancing the development of its own web search service and is expected to receive technical support from Microsoft’s Bing search engine. Although the absence of GPT-5 and AI search engines may disappoint some viewers, the upgraded content of ChatGPT and its possible future plans are still full of expectations.

Posted on Leave a comment

Experts advise against purchasing low-cost air tickets: they are often “black tickets”

Some consumers complained that they were cheated by air tickets during the past May Day holiday. They booked tickets in advance at a lower price through a “ticket agent” on a second-hand trading platform, but were told that the ticket information could not be found when they arrived at the boarding gate with their whole family.

The seller stopped responding on the grounds of closing the order and refunding the money, and ultimately had to buy the ticket at full price to avoid delaying the trip.

It is reported that this is a typical “black ticket agent”, who does not have the qualification of ticket sales agent, and basically relies on mileage points or credit card fraud to obtain so-called “special air tickets”, and then cheats consumers under the banner of low prices.

Once someone purchases a ticket through them, they may either be stuck in a long and anxious wait, or be deceived by the other party’s fake itinerary using Photoshop, and may also lose money and reveal personal privacy.

Posted on Leave a comment

2024新款金属皮带扣鳄鱼皮凉鞋

Note:2024新款金属皮带扣鳄鱼皮凉鞋女性感黑丝方头一字扣带细跟高跟鞋。

品牌:Chaeosni Shwaizi 尺码:34,35,36,37,38,39,40 风格:欧美

流行元素:浅口 后跟高:高跟(5-8cm)

颜色分类:鳄鱼纹【7cm】,鳄鱼纹【9cm】,黑色【7cm】,黑色【9cm】

上市年份季节:2024年春季 鞋头款式:包头 跟底款式:细跟

里料材质:羊皮 适用对象:青年(18-40周岁) 后帮:后绊带

款式:一字式扣带 帮面材质:鳄鱼皮

2024新款金属皮带扣鳄鱼皮凉鞋女性感黑丝方头一字扣带细跟高跟鞋-淘宝网 (taobao.com)