[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-74002":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":23,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":27,"readmeContent":28,"aiSummary":29,"trendingCount":16,"starSnapshotCount":16,"syncStatus":30,"lastSyncTime":31,"discoverSource":32},74002,"AI-Scientist","SakanaAI\u002FAI-Scientist","SakanaAI","The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery 🧑‍🔬","",null,"Jupyter Notebook",13941,1981,160,91,0,41,92,379,123,119.89,"Other",false,"main",true,[],"2026-06-12 04:01:12","\u003Ch1 align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fblob\u002Fmain\u002Fdocs\u002Flogo_2.png\">\n    \u003Cimg src=\"docs\u002Flogo_2.png\" width=\"215\" \u002F>\u003C\u002Fa>\u003Cbr>\n  \u003Cb>The AI Scientist: Towards Fully Automated\u003C\u002Fb>\u003Cbr>\n  \u003Cb>Open-Ended Scientific Discovery 🧑‍🔬\u003C\u002Fb>\u003Cbr>\n\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n  📚 \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.06292\">[Paper]\u003C\u002Fa> |\n  📝 \u003Ca href=\"https:\u002F\u002Fsakana.ai\u002Fai-scientist\u002F\">[Blog Post]\u003C\u002Fa> |\n  📂 \u003Ca href=\"https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1G7A0wTqfXVa-cpexjk0oaXakaSJwffEt\">[Drive Folder]\u003C\u002Fa>\n\u003C\u002Fp>\n\nOne of the grand challenges of artificial intelligence is developing agents capable of conducting scientific research and discovering new knowledge. While frontier models have already been used to aid human scientists—for example, for brainstorming ideas or writing code—they still require extensive manual supervision or are heavily constrained to specific tasks.\n\nWe're excited to introduce **The AI Scientist**, the first comprehensive system for fully automatic scientific discovery, enabling Foundation Models such as Large Language Models (LLMs) to perform research independently.\n\nWe provide all runs and data from our paper [here](https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1G7A0wTqfXVa-cpexjk0oaXakaSJwffEt?usp=sharing), where we run each base model on each template for approximately 50 ideas. We *highly* recommend reading through some of the [Claude papers](https:\u002F\u002Fdrive.google.com\u002Fdrive\u002Ffolders\u002F1Mmpz6M1FK4q8e-SewgZcUzdeD0Q2zC39?usp=sharing) to get a sense of the system's strengths and weaknesses. Here are some example papers generated by **The AI Scientist** 📝:\n\n1. [DualScale Diffusion: Adaptive Feature Balancing for Low-Dimensional Generative Models](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fblob\u002Fmain\u002Fexample_papers\u002Fadaptive_dual_scale_denoising.pdf)\n2. [Multi-scale Grid Noise Adaptation: Enhancing Diffusion Models For Low-dimensional Data](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fblob\u002Fmain\u002Fexample_papers\u002Fgrid_based_noise_adaptation.pdf)\n3. [GAN-Enhanced Diffusion: Boosting Sample Quality and Diversity](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fblob\u002Fmain\u002Fexample_papers\u002Fgan_diffusion.pdf)\n4. [DualDiff: Enhancing Mode Capture in Low-dimensional Diffusion Models via Dual-expert Denoising](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Ftree\u002Fmain\u002Fexample_papers\u002Fdual_expert_denoiser.pdf) \n5. [StyleFusion: Adaptive Multi-style Generation in Character-Level Language Models](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fblob\u002Fmain\u002Fexample_papers\u002Fmulti_style_adapter.pdf)\n6. [Adaptive Learning Rates for Transformers via Q-Learning](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Ftree\u002Fmain\u002Fexample_papers\u002Frl_lr_adaptation.pdf)\n7. [Unlocking Grokking: A Comparative Study of Weight Initialization Strategies in Transformer Models](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Ftree\u002Fmain\u002Fexample_papers\u002Fweight_initialization_grokking.pdf)\n8. [Grokking Accelerated: Layer-wise Learning Rates for Transformer Generalization](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Ftree\u002Fmain\u002Fexample_papers\u002Flayerwise_lr_grokking.pdf)\n9. [Grokking Through Compression: Unveiling Sudden Generalization via Minimal Description Length](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Ftree\u002Fmain\u002Fexample_papers\u002Fmdl_grokking_correlation.pdf)\n10. [Accelerating Mathematical Insight: Boosting Grokking Through Strategic Data Augmentation](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Ftree\u002Fmain\u002Fexample_papers\u002Fdata_augmentation_grokking.pdf)\n\n> **Note:**  \n> **Caution!** This codebase will execute LLM-written code. There are various risks and challenges associated with this autonomy, including the use of potentially dangerous packages, web access, and potential spawning of processes. Use at your own discretion. Please make sure to [containerize](#containerization) and restrict web access appropriately.\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fblob\u002Fmain\u002Fexample_papers\u002Fadaptive_dual_scale_denoising\u002Fadaptive_dual_scale_denoising.pdf\">\u003Cimg src=\"https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fblob\u002Fmain\u002Fdocs\u002Fanim-ai-scientist.gif\" alt=\"Adaptive Dual Scale Denoising\" width=\"80%\" \u002F>\n\u003C\u002Fa>\u003C\u002Fp>\n\n## Table of Contents\n\n1. [Introduction](#introduction)\n2. [Requirements](#requirements)\n   - [Installation](#installation)\n   - [Supported Models and API Keys](#supported-models-and-api-keys)\n3. [Setting Up the Templates](#setting-up-the-templates)\n   - [NanoGPT Template](#nanogpt-template)\n   - [2D Diffusion Template](#2d-diffusion-template)\n   - [Grokking Template](#grokking-template)\n4. [Run AI Scientist Paper Generation Experiments](#run-ai-scientist-paper-generation-experiments)\n5. [Getting an LLM-Generated Paper Review](#getting-an-llm-generated-paper-review)\n6. [Making Your Own Template](#making-your-own-template)\n   - [Community-Contributed Templates](#community-contributed-templates)\n7. [Template Resources](#template-resources)\n8. [Citing The AI Scientist](#citing-the-ai-scientist)\n9. [Frequently Asked Questions](#frequently-asked-questions)\n10. [Containerization](#containerization)\n\n## Introduction\n\nWe provide three templates, which were used in our paper, covering the following domains: **NanoGPT**, **2D Diffusion**, and **Grokking**. These templates enable The AI Scientist to generate ideas and conduct experiments in these areas. We accept contributions of new templates from the community, but please note that they are not maintained by us. All other templates beyond the three provided are community contributions.\n\n## Requirements\n\nThis code is designed to run on Linux with NVIDIA GPUs using CUDA and PyTorch. Support for other GPU architectures may be possible by following the [PyTorch guidelines](https:\u002F\u002Fpytorch.org\u002Fget-started\u002Flocally\u002F). The current templates would likely take an infeasible amount of time on CPU-only machines. Running on other operating systems may require significant adjustments.\n\n### Installation\n\n```bash\nconda create -n ai_scientist python=3.11\nconda activate ai_scientist\n# Install pdflatex\nsudo apt-get install texlive-full\n\n# Install PyPI requirements\npip install -r requirements.txt\n```\n\n**Note:** Installing `texlive-full` can take a long time. You may need to [hold Enter](https:\u002F\u002Faskubuntu.com\u002Fquestions\u002F956006\u002Fpregenerating-context-markiv-format-this-may-take-some-time-takes-forever) during the installation.\n\n### Supported Models and API Keys\n\nWe support a wide variety of models, including open-weight and API-only models. In general, we recommend using only frontier models above the capability of the original GPT-4. To see a full list of supported models, see [here](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fblob\u002Fmain\u002Fai_scientist\u002Fllm.py).\n\n#### OpenAI API (GPT-4o, GPT-4o-mini, o1 models)\n\nBy default, this uses the `OPENAI_API_KEY` environment variable.\n\n#### Anthropic API (Claude Sonnet 3.5)\n\nBy default, this uses the `ANTHROPIC_API_KEY` environment variable.\n\n##### Claude Models via Bedrock\n\nFor Claude models provided by [Amazon Bedrock](https:\u002F\u002Faws.amazon.com\u002Fbedrock\u002F), please install these additional packages:\n\n```bash\npip install anthropic[bedrock]\n```\n\nNext, specify a set of valid [AWS Credentials](https:\u002F\u002Fdocs.aws.amazon.com\u002Fcli\u002Fv1\u002Fuserguide\u002Fcli-configure-envvars.html) and the target [AWS Region](https:\u002F\u002Fdocs.aws.amazon.com\u002Fbedrock\u002Flatest\u002Fuserguide\u002Fbedrock-regions.html):\n\nSet the environment variables: `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, `AWS_REGION_NAME`.\n\n##### Claude Models via Vertex AI\n\nFor Claude models provided by [Vertex AI Model Garden](https:\u002F\u002Fcloud.google.com\u002Fmodel-garden?hl=en), please install these additional packages:\n\n```bash\npip install google-cloud-aiplatform\npip install anthropic[vertex]\n```\n\nNext, set up valid authentication for a [Google Cloud project](https:\u002F\u002Fcloud.google.com\u002Fvertex-ai\u002Fdocs\u002Fauthentication), for example by providing the region and project ID:\n\n```bash\nexport CLOUD_ML_REGION=\"REGION\"           # for Model Garden call\nexport ANTHROPIC_VERTEX_PROJECT_ID=\"PROJECT_ID\"  # for Model Garden call\nexport VERTEXAI_LOCATION=\"REGION\"         # for Aider\u002FLiteLLM call\nexport VERTEXAI_PROJECT=\"PROJECT_ID\"      # for Aider\u002FLiteLLM call\n```\n\n#### DeepSeek API (deepseek-chat, deepseek-reasoner)\nBy default, this uses the `DEEPSEEK_API_KEY` environment variable.\n\n#### OpenRouter API (Llama3.1)\n\nBy default, this uses the `OPENROUTER_API_KEY` environment variable.\n\n#### Google Gemini\nWe support Google Gemini models (e.g., \"gemini-1.5-flash\", \"gemini-1.5-pro\") via the [google-generativeai](https:\u002F\u002Fpypi.org\u002Fproject\u002Fgoogle-generativeai) Python library. By default, it uses the environment variable:\n\n```bash\nexport GEMINI_API_KEY=\"YOUR GEMINI API KEY\"\n```\n\n#### Semantic Scholar API (Literature Search)\n\nOur code can also optionally use a Semantic Scholar API Key (`S2_API_KEY`) for higher throughput [if you have one](https:\u002F\u002Fwww.semanticscholar.org\u002Fproduct\u002Fapi), though it should work without it in principle. If you have problems with Semantic Scholar, you can skip the literature search and citation phases of paper generation.\n\nBe sure to provide the key for the model used for your runs, e.g.:\n\n```bash\nexport OPENAI_API_KEY=\"YOUR KEY HERE\"\nexport S2_API_KEY=\"YOUR KEY HERE\"\n```\n\n#### OpenAlex API (Literature Search Alternative)\n\nOpenAlex API can be used as an alternative if you do not have a Semantic Scholar API Key.\nOpenAlex does not require API key.\n\n```bash\npip install pyalex\nexport OPENALEX_MAIL_ADDRESS=\"YOUR EMAIL ADDRESS\"\n```\n\nAnd specify `--engine openalex` when you execute the AI Scientist code.\n\nNote that this is experimental for those who do not have a Semantic Scholar API Key.\n\n## Setting Up the Templates\n\nThis section provides instructions for setting up each of the three templates used in our paper. Before running The AI Scientist experiments, please ensure you have completed the setup steps for the templates you are interested in.\n\n### NanoGPT Template\n\n**Description:** This template investigates transformer-based autoregressive next-token prediction tasks.\n\n**Setup Steps:**\n\n1. **Prepare the data:**\n\n   ```bash\n   python data\u002Fenwik8\u002Fprepare.py\n   python data\u002Fshakespeare_char\u002Fprepare.py\n   python data\u002Ftext8\u002Fprepare.py\n   ```\n\n2. **Create baseline runs (machine dependent):**\n\n   ```bash\n   # Set up NanoGPT baseline run\n   # NOTE: YOU MUST FIRST RUN THE PREPARE SCRIPTS ABOVE!\n   cd templates\u002FnanoGPT\n   python experiment.py --out_dir run_0\n   python plot.py\n   ```\n\n### 2D Diffusion Template\n\n**Description:** This template studies improving the performance of diffusion generative models on low-dimensional datasets.\n\n**Setup Steps:**\n\n1. **Install dependencies:**\n\n   ```bash\n   # Set up 2D Diffusion\n   git clone https:\u002F\u002Fgithub.com\u002Fgregversteeg\u002FNPEET.git\n   cd NPEET\n   pip install .\n   pip install scikit-learn\n   ```\n\n2. **Create baseline runs:**\n\n   ```bash\n   # Set up 2D Diffusion baseline run\n   cd templates\u002F2d_diffusion\n   python experiment.py --out_dir run_0\n   python plot.py\n   ```\n\n### Grokking Template\n\n**Description:** This template investigates questions about generalization and learning speed in deep neural networks.\n\n**Setup Steps:**\n\n1. **Install dependencies:**\n\n   ```bash\n   # Set up Grokking\n   pip install einops\n   ```\n\n2. **Create baseline runs:**\n\n   ```bash\n   # Set up Grokking baseline run\n   cd templates\u002Fgrokking\n   python experiment.py --out_dir run_0\n   python plot.py\n   ```\n\n## Run AI Scientist Paper Generation Experiments\n\n**Note:** Please ensure the setup steps above are completed before running these experiments.\n\n```bash\nconda activate ai_scientist\n# Run the paper generation.\npython launch_scientist.py --model \"gpt-4o-2024-05-13\" --experiment nanoGPT_lite --num-ideas 2\npython launch_scientist.py --model \"claude-3-5-sonnet-20241022\" --experiment nanoGPT_lite --num-ideas 2\n```\n\nIf you have more than one GPU, use the `--parallel` option to parallelize ideas across multiple GPUs.\n\n## Getting an LLM-Generated Paper Review\n\n```python\nimport openai\nfrom ai_scientist.perform_review import load_paper, perform_review\n\nclient = openai.OpenAI()\nmodel = \"gpt-4o-2024-05-13\"\n\n# Load paper from PDF file (raw text)\npaper_txt = load_paper(\"report.pdf\")\n\n# Get the review dictionary\nreview = perform_review(\n    paper_txt,\n    model,\n    client,\n    num_reflections=5,\n    num_fs_examples=1,\n    num_reviews_ensemble=5,\n    temperature=0.1,\n)\n\n# Inspect review results\nreview[\"Overall\"]    # Overall score (1-10)\nreview[\"Decision\"]   # 'Accept' or 'Reject'\nreview[\"Weaknesses\"] # List of weaknesses (strings)\n```\n\nTo run batch analysis:\n\n```bash\ncd review_iclr_bench\npython iclr_analysis.py --num_reviews 500 --batch_size 100 --num_fs_examples 1 --num_reflections 5 --temperature 0.1 --num_reviews_ensemble 5\n```\n\n## Making Your Own Template\n\nIf there is an area of study you would like **The AI Scientist** to explore, it is straightforward to create your own templates. In general, follow the structure of the existing templates, which consist of:\n\n- `experiment.py` — This is the main script where the core content is. It takes an argument `--out_dir`, which specifies where it should create the folder and save the relevant information from the run.\n- `plot.py` — This script takes the information from the `run` folders and creates plots. The code should be clear and easy to edit.\n- `prompt.json` — Put information about your template here.\n- `seed_ideas.json` — Place example ideas here. You can also try to generate ideas without any examples and then pick the best one or two to put here.\n- `latex\u002Ftemplate.tex` — We recommend using our LaTeX folder but be sure to replace the pre-loaded citations with ones that you expect to be more relevant.\n\nThe key to making new templates work is matching the base filenames and output JSONs to the existing format; everything else is free to change.\nYou should also ensure that the `template.tex` file is updated to use the correct citation style \u002F base plots for your template.\n\n### Community-Contributed Templates\n\nWe welcome community contributions in the form of new templates. While these are not maintained by us, we are delighted to highlight your templates to others. Below, we list community-contributed templates along with links to their pull requests (PRs):\n\n- Infectious Disease Modeling (`seir`) - [PR #137](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fpull\u002F137)\n- Image Classification with MobileNetV3 (`mobilenetV3`) - [PR #141](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fpull\u002F141)\n- Sketch RNN (`sketch_rnn`) - [PR #143](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fpull\u002F143)\n- AI in Quantum Chemistry (`MACE`) - [PR#157](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fpull\u002F157)\n- Earthquake Prediction (`earthquake-prediction`) - [PR #167](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fpull\u002F167)\n- Tensorial Radiance Fields (`tensorf`) - [PR #175](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fpull\u002F175)\n- Large Language Model Steering \u002F Probes (`probes`) - [PR #215](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fpull\u002F215)\n\n*This section is reserved for community contributions. Please submit a pull request to add your template to the list! Please describe the template in the PR description, and also show examples of the generated papers.*\n\n## Template Resources\n\nWe provide three templates, which heavily use code from other repositories, credited below:\n\n- **NanoGPT Template** uses code from [NanoGPT](https:\u002F\u002Fgithub.com\u002Fkarpathy\u002FnanoGPT) and this [PR](https:\u002F\u002Fgithub.com\u002Fkarpathy\u002FnanoGPT\u002Fpull\u002F254).\n- **2D Diffusion Template** uses code from [tiny-diffusion](https:\u002F\u002Fgithub.com\u002Ftanelp\u002Ftiny-diffusion), [ema-pytorch](https:\u002F\u002Fgithub.com\u002Flucidrains\u002Fema-pytorch), and [Datasaur](https:\u002F\u002Fwww.research.autodesk.com\u002Fpublications\u002Fsame-stats-different-graphs\u002F).\n- **Grokking Template** uses code from [Sea-Snell\u002Fgrokking](https:\u002F\u002Fgithub.com\u002FSea-Snell\u002Fgrokking) and [danielmamay\u002Fgrokking](https:\u002F\u002Fgithub.com\u002Fdanielmamay\u002Fgrokking).\n\nWe would like to thank the developers of the open-source models and packages for their contributions and for making their work available.\n\n## Citing The AI Scientist\n\nIf you use **The AI Scientist** in your research, please cite it as follows:\n\n```\n@article{lu2024aiscientist,\n  title={The {AI} {S}cientist: Towards Fully Automated Open-Ended Scientific Discovery},\n  author={Lu, Chris and Lu, Cong and Lange, Robert Tjarko and Foerster, Jakob and Clune, Jeff and Ha, David},\n  journal={arXiv preprint arXiv:2408.06292},\n  year={2024}\n}\n```\n\n## Frequently Asked Questions\n\nWe recommend reading our paper first for any questions you have on The AI Scientist.\n\n**Why am I missing files when running The AI Scientist?**\n\nEnsure you have completed all the setup and preparation steps before the main experiment script.\n\n**Why has a PDF or a review not been generated?**\n\nThe AI Scientist finishes an idea with a success rate that depends on the template, the base foundation model, and the complexity of the idea. We advise referring to our main paper. The highest success rates are observed with Claude Sonnet 3.5. Reviews are best done with GPT-4o; all other models have issues with positivity bias or failure to conform to required outputs.\n\n**What is the cost of each idea generated?**\n\nTypically less than $15 per paper with Claude Sonnet 3.5. We recommend DeepSeek Coder V2 for a much more cost-effective approach. A good place to look for new models is the [Aider leaderboard](https:\u002F\u002Faider.chat\u002Fdocs\u002Fleaderboards\u002F).\n\n**How do I change the base conference format associated with the write-ups?**\n\nChange the base `template.tex` files contained within each template.\n\n**How do I run The AI Scientist for different subject fields?**\n\nPlease refer to the instructions for different templates. In this current iteration, this is restricted to ideas that can be expressed in code. However, lifting this restriction would represent exciting future work! :)\n\n**How do I add support for a new foundation model?**\n\nYou may modify `ai_scientist\u002Fllm.py` to add support for a new foundation model. We do not advise using any model that is significantly weaker than GPT-4 level for **The AI Scientist**.\n\n**Why do I need to run the baseline runs myself?**\n\nThese appear as `run_0` and should be run per machine you execute **The AI Scientist** on for accurate run-time comparisons due to hardware differences.\n\n**What if I have problems accessing the Semantic Scholar API?**\n\nWe use the Semantic Scholar API to check ideas for novelty and collect citations for the paper write-up. You may be able to skip these phases if you don't have an API key or the API is slow to access.\n\n## Containerization\n\nWe include a [community-contributed](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist\u002Fpull\u002F21) Docker image that may assist with your containerization efforts in `experimental\u002FDockerfile`.\n\nYou can use this image like this:\n\n```bash\n# Endpoint Script\ndocker run -e OPENAI_API_KEY=$OPENAI_API_KEY -v `pwd`\u002Ftemplates:\u002Fapp\u002FAI-Scientist\u002Ftemplates \u003CAI_SCIENTIST_IMAGE> \\\n  --model gpt-4o-2024-05-13 \\\n  --experiment 2d_diffusion \\\n  --num-ideas 2\n```\n\n```bash\n# Interactive\ndocker run -it -e OPENAI_API_KEY=$OPENAI_API_KEY \\\n  --entrypoint \u002Fbin\u002Fbash \\\n  \u003CAI_SCIENTIST_IMAGE>\n```\n\n## ⚖️ License & Responsible Use\n\nThis project is licensed under **The AI Scientist Source Code License** (a derivative of the Responsible AI License). \n\n**Mandatory Disclosure:** By using this code, you are legally bound to clearly and prominently disclose the use of AI in any resulting scientific manuscripts or papers. \n\nWe recommend the following attribution in your paper's Abstract or Methods section:\n> \"This manuscript was autonomously generated using [The AI Scientist](https:\u002F\u002Fgithub.com\u002FSakanaAI\u002FAI-Scientist).\"\n\n## Star History\n\n[![Star History Chart](https:\u002F\u002Fapi.star-history.com\u002Fsvg?repos=SakanaAI\u002FAI-Scientist&type=Date)](https:\u002F\u002Fstar-history.com\u002F#SakanaAI\u002FAI-Scientist&Date)\n","AI Scientist 项目致力于实现全自动化的开放性科学研究。它利用大型语言模型等基础模型独立进行科研工作，生成新的科学想法和论文。该项目的核心功能包括自动提出研究假设、设计实验方案以及撰写科研文章，显著减少了人工干预的需求。技术上，它通过特定的模板指导模型产生创意，并能够处理多种类型的科学研究任务。适用于需要创新思维的研究领域，特别是那些希望探索由AI驱动的新知识发现过程的场景。",2,"2026-06-11 03:48:21","high_star"]