Hello!

Every week, there have been remarkable advances in artificial intelligence, including more powerful, versatile and accurate large language models for daily applications. Indeed, a recent post discussed some of the use cases of such tools in research and their benefits in accelerating software development or improving scientific content writing. In this post, I would like to focus on some of the tools I use to benefit from these advancements.

1. General support tools: Perplexity, Poe, Pi

2. Software Development: GitHub Copilot, Perplexity Labs, Cursor

3. Academic research: ChatPDF, Claude, Grammarly

General support tools: Perplexity, Poe, Pi

These are freely available tools and websites where users can directly provide questions to be answered. Depending on the use, answers and the input length can vary, too. Nonetheless, these have proved extremely valuable for accessing quick information or getting support on simple questions in daily life.

Conveniently, Poe includes several bots built on well-established models from different entities, providing immediate access to assistants specialised for various purposes and tasks. Nonetheless, non-subscribers have a limited number of credits to spend daily, which can restrict the length of the interactions. Besides, as different bots cost different credits, one may also need to carefully manage their balance appropriately. In any case, I’ve found it very versatile and convenient, and it provides sufficient credits for my daily usage.

On the other hand, Pi behaves more like a human, with a very friendly discourse that still contains all the information in its answers and a clean interface that focuses on the conversation without distractions. To my knowledge, it does not limit the number of interactions but also does not provide the ability to start parallel conversations for different topics (unless we open multiple browser tabs). Instead, we have a single page with the entire conversation. Finally, Perplexity is another platform available for free or with a subscription. Still, the most noteworthy aspect for me is its care in providing sources for sections of the answers, such that the user can conveniently access the original website and obtain the complete context that was isolated and summarised by the bot.

Importantly, Poe and Pi have a mobile version, which dramatically increases their usability in most circumstances, allowing users to quickly prompt questions on the go.

Software Development: GitHub Copilot, Perplexity Labs, Cursor

For work purposes, one of the main tools available is GitHub Copilot. This is a Visual Studio Code extension, and I’ve been using it for a few months and have observed some improvements as the developers update the service. I am particularly excited about the auto-completion feature, which provides meaningful and often accurate suggestions based on open scripts. If you have an educational email (from a university or institution), you can benefit from the GitHub Education Pack, which provides free access to GitHub Copilot. While it offers many more features, including a chat where we can ask specific questions or get support to explain and debug code, I have not tried these extensively. Nonetheless, I am pleased with code suggestions for simple tasks like typing or its ability to instantly suggest the documentation of functions based on their body.

Additionally, for cases where I am interested in learning how to optimise or refactor code, I have also experimented with Perplexity Labs, an alternative service provided by Perplexity where users can try the latest models available, many of which were designed for software-related tasks. For instance, I often use codellama-70b-instruct to understand new code or ask for optimisations. Overall, these models provide friendly and insightful answers that have supported me in writing clean, efficient code.

At last, I have also been testing Cursor, a code editor based on Visual Studio Code that seamlessly integrates AI tools like GPT models. Users with access to API keys can set them to enable chatting and software development with hints from the GPT model selected. It is straightforward to highlight a few lines of code and ask for clarification or directly request an edit based on specific instructions. More importantly, Cursor allows users to add links to custom documentation and API pages, which can be later mentioned when asking questions in the chat, so the model will use this information to support its answer. Hence, when learning new tools, it is possible to ask how to implement a specific function and receive support grounded on the documentation available, making the learning curve much easier and more convenient.

Academic research: ChatPDF, Claude, Grammarly

Finally, there have been many services targeting the scientific community. For such tasks, I am still looking for a free and impactful system. Depending on the task, I have tried ChatPDF and PDFPals, which help quickly obtain answers from PDFs and are extremely handy when looking for details in long research papers. Alternatively, I usually use Claude, available on Poe, which I have found to provide the most natural writing suggestions for writing tasks. Typically, I am interested in editing or writing small sections of text, such that the context window available for the models provided in these services is sufficient for me to explain what I am trying to achieve and provide my first draft as input to be rewritten. I am also aware that Grammarly has introduced AI-supported tools on its platform, but I still need to try them out. I am a Grammarly subscriber, and I find the suggestions helpful, so I am excited to learn how they improved their service with AI models. Indeed, I have yet to have a long and demanding writing task that could allow me to explore AI-driven writing assistants in depth. Other than that, I have also experimented with scite’s assistant to search for relevant literature based on specific requests, but I was using an early version of the service which, in my opinion, was still not able to provide the most up-to-date research depending on the prompt. Nonetheless, the effort and care to provide references for relevant statements in the answer, similar to Perplexity, is worth praising.

Conclusion

With the rapid development of novel, more powerful large language models, I suspect these will become ubiquitous in daily life and likely a part of many jobs, including academic research. So far, I am excited to experiment with such tools and understand how they can support my work positively and help me become more productive. Nonetheless, I also acknowledge the copyright questions that many of these systems raise, and I aim to be aware of existing regulations (e.g., scientific journal guidelines) that limit or ban the use of some tools. For now, we can continue admiring this rapid development of technologies and provide our feedback to developers (e.g., on social media) to help steer new advances towards fruitful goals that are also ethically acceptable.

Please feel free to share your thoughts on AI-tools for scientific research!

Have a great day!

Acknowledgements

I have no affiliation with any of the brands or products discussed in this article. All opinions are my own.