Generative AI – encompassing models like large language models (LLMs), diffusion models, and Variational Autoencoders (VAEs), – are transforming how scientists approach discovery. Unlike traditional AI which recognizes patterns or makes predictions, generative models create new data or designs. In scientific research, this ability unlocks novel hypotheses, accelerates simulations, and produces synthetic data that drives innovation.
Table. Examples of research uses of generative AI.
| GenAI Method | Use | Domain |
|---|---|---|
| LLM: Geneformer | Disease modeling and identifying therapeutic targets | Biology |
| LLM: GPT-4 | Autonomous chemical research | Chemistry |
| LLM: ChemCrow | Chemical Synthesis | Chemistry |
| LLM: GPT-4 | Computational simulation of biological processes | Biology |
| LLM: GPT-4 | VR-based human-robot teaming simulation | Robotics |
New ‘deep research’ models for professional productivity
In recent years, large language models (LLMs) like ChatGPT have transformed the research process by enabling fast, natural-language responses to complex questions. However, early LLMs had significant limitations regarding academic research — they generated plausible but often inaccurate information, could not verify sources, and couldn’t provide real-time access to up-to-date research. This made it difficult for researchers to fully trust or rely on the outputs for academic work.
Deep research models represent a significant improvement over traditional LLMs. Tools like Perplexity, Consensus, and OpenRead combine the power of language models with real-time search capabilities and citation tracking. This means they can:
- Search for and retrieve real-time information from the web and academic databases.
- Provide direct citations for generated statements, enabling researchers to verify the sources and trace the origin of the information.
This shift enhances the trustworthiness and transparency of AI-generated content, making it more suitable for academic and professional research. However, it’s important to note that AI-generated outputs, regardless of how advanced the models are, should be treated as starting points rather than final products. Researchers should avoid directly copying and pasting AI-generated text into their work. Instead, they should use these tools to identify key information, gather insights, and guide their critical thinking and further research. AI research tools are best viewed as research assistants — not replacements for scholarly rigor and analytical reasoning.
Table. Example ‘deep research’ models for literature search, summary, and idea generation.
| Model | Description |
|---|---|
| OpenAI Deep Research | A large language model (LLM) developed by OpenAI based on the GPT (Generative Pre-trained Transformer) architecture. It generates human-like text, answers questions, and can assist with various language tasks. |
| Perplexity Deep Research | AI-powered search engine that combines language model capabilities with real-time web access to provide accurate and sourced answers. |
| Consensus | AI-powered research tool that searches academic papers and synthesizes answers from peer-reviewed sources. Focused on providing evidence-based answers. |
| OpenRead | AI research tool focused on summarizing and analyzing scientific papers. It includes features like citation tracking and related paper suggestions. |
| Elicit | AI tool for automating research tasks like finding papers, extracting key points, and generating summaries. |
| Scite | AI-powered citation and paper analysis tool that shows how papers have been cited (supporting or disputing). |
| ResearchRabbit | A research tool that creates a visual map of papers and citations, helping researchers discover related works and trends. |
| SciSpace (formerly Typeset) | An AI-powered platform for reading, annotating, and summarizing research papers. Also provides explanations for technical terms. |
| Litmaps | Citation-based tool that builds research maps to visualize connections between papers and identify research gaps. |
Analytical Methods
Advances in AI have not only transformed academic research but also enhanced general research workflows. Beyond language models designed for text-based research, a growing ecosystem of generative AI-powered tools can now assist with coding, data analysis, and visualization. These tools provide researchers with powerful capabilities to automate complex processes, improve productivity, and explore new creative directions.
AI research tools for coding and visualization, such as Claude and Napkin AI, leverage advanced natural language processing and generative AI to simplify complex tasks. They allow researchers to write and debug code more efficiently, generate high-quality figures and plots from data, and create structured outputs with minimal manual effort.
Table. Generative AI for data analysis and visualization
| Model | Description |
|---|---|
| Napkin AI | AI-based figure creation tool that helps researchers generate high-quality charts, diagrams, and visual summaries from data and text input. |
| GitHub Copilot | AI coding assistant integrated with GitHub that suggests code completions and helps debug code. |
| Snorkel | AI-based tool for labeling datasets and training machine learning models with weak supervision. |