INTENTION
The purpose of this article is to share my experience attempting to implement GPT-Researcher on my M1 silicon-based MacBook Pro. I’ll provide an overview of the process, including my struggles, lessons learned, and high-level reflections on AI agents. This article also includes links to helpful resources, pain points I encountered, and tips for navigating the complexities of implementing such tools. In collaboration with OpenAI’s ChatGPT, this narrative has been refined to ensure clarity and accessibility.
Before diving in, let me give you some context.
CONTEXT
I am a graphic designer with 20 years of experience, as well as a design educator pursuing a Master’s in Digital Management at Teesside University via Hyper Island in Sweden.
Currently, I am studying a module entitled Intelligent Machines, which explores trends and technologies like artificial intelligence (AI), machine learning (ML), large language models (LLMs), quantum computing, neural networks, and edge computing. As part of this module, we investigate how future technologies can support individuals with disabilities in finding meaningful employment.
The rapid pace of innovation in AI often feels overwhelming; what appears on the horizon today may evolve drastically tomorrow. My research has sparked a fascination with the concept of agency in artificial intelligence—how AI can act autonomously to support human objectives. Motivated by curiosity, I decided to experiment with implementing an AI agent to assist with my research. While I hoped for a seamless experience, the reality proved far more challenging.
However, the insights gained from this exploration will be invaluable in writing my research paper, where I aim to evaluate user-centered AI solutions. Inspired by the work of Professor James Landay at Stanford HAI, I am exploring how AI can prioritize accessibility and inclusivity in design. This work directly supports our ongoing project with Samhall.se, where we are developing AI-driven tools to empower individuals with disabilities and enhance operational efficiency.
WORKING WITH GPT-RESEARCHER
Initial Steps
A quick search using ChatGPT pointed me to the GitHub repository for GPT-Researcher. I reviewed several YouTube tutorials on installing the program, which seemed straightforward. However, as someone with limited coding experience—mostly some HTML from my design background—diving into terminal commands felt daunting.
Installation Struggles
Implementing GPT-Researcher required patience, a foundational understanding of coding, and familiarity with file structures and operating systems. Over two days, I wrestled with installing GPT-Researcher, repeatedly installing and uninstalling the program. After about five hours of trial and error, I switched gears to include ChatGPT in my workflow.
By copying terminal outputs into ChatGPT and receiving guidance, I managed to debug Python scripts and navigate installation errors. This collaboration proved instrumental in eventually getting the program up and running.
Running the Agent
Even with the agent installed, new hurdles arose. The program encountered errors when attempting to use Docker, and I discovered I needed to purchase OpenAI tokens for the agent to function. After resolving the payment issue, I initiated a research task, but it was repeatedly canceled due to excessive token consumption, which triggered OpenAI’s rate limits.
I spent hours tweaking the code (with ChatGPT’s help) to throttle token usage but ultimately failed. After this setback, I uninstalled and reinstalled GPT-Researcher, now more confident with the installation process. Despite following new instructional videos, I still encountered runtime errors when trying to access a directory of research PDFs.
Final Attempts
Eventually, I succeeded in installing a functional version of the program using additional API keys. However, the process remained tedious and opaque. While I have yet to fully evaluate its utility in my research, I remain cautious about relying on it exclusively. As I work toward my Master’s degree, ethical considerations require me to critically synthesize any AI-provided insights.
LINKS TO RESOURCES
Here are some useful links for those attempting to implement GPT-Researcher:
“The rapid pace of innovation in AI often feels overwhelming; what appears on the horizon today may evolve drastically tomorrow.”
PAIN POINTS
- Ambiguity in Instructions: The lack of clear, step-by-step guidance posed significant challenges for someone new to coding.
- Complex Fee Structure: Navigating payments across multiple platforms with limited transparency about costs was confusing.
- Limited Functionality: The agent lacks interactive feedback loops, making prompt refinement cumbersome.
- Clunky Interface: The user interface is unintuitive, and editing prompts is unnecessarily difficult.
- Runtime Errors: Persistent issues with file directories and token management hindered progress.
TIPS FOR USING GPT-RESEARCHER
-
Limit Document Size and Volume: When asking GPT-Researcher to process research from your local documents in the My-docs directory, keep the size and number of files minimal. This can prevent the agent from overloading the token requests to your LLM API. Learn more about managing API usage and token limits.
-
Understand Token Usage: Familiarize yourself with OpenAI’s token consumption rules to avoid unexpected cancellations. Setting realistic limits in the agent’s configuration can help manage API interactions more effectively. Refer to OpenAI’s token consumption guide.
-
Follow Reliable Installation Guides: Use comprehensive tutorials tailored to your operating system, such as those specifically designed for M1 Macs. Consider referencing Docker’s installation guide, Homebrew’s official site, and the Tavily API for streamlined integration options.
-
Use GPT Collaboratively: If you’re struggling with technical challenges, leverage ChatGPT to debug errors by providing detailed terminal outputs. Iterative collaboration with ChatGPT can help overcome roadblocks and improve installation efficiency.
-
Keep API Keys Secure and Updated: Ensure both primary and secondary API keys are correctly configured and active, as missing keys often cause runtime errors. Manage your keys effectively through the OpenAI API dashboard.
REFLECTIONS
The challenges I faced highlight a critical issue with emerging AI technologies: accessibility. The technical barriers to implementing tools like GPT-Researcher render them inaccessible to the average user. Until these tools are supported by more user-friendly interfaces, they will remain the domain of tech-savvy individuals.
This divide raises ethical concerns. If only a subset of people can harness the power of AI agents, the technology may exacerbate existing inequalities. However, the insights I gained will contribute to designing more inclusive AI solutions, with a focus on empowering individuals with disabilities and improving operational efficiency for organizations like Samhall.se.
While the concept of AI agents is exciting, their current limitations in usability, interactivity, and transparency hinder widespread adoption. I remain optimistic, however, that future iterations will address these issues, enabling broader use of AI agents to improve workflows across industries.
CONCLUSION
My experience with GPT-Researcher was both frustrating and enlightening. It underscored the importance of user-centric design in emerging technologies and the need for transparency in payment models. As AI tools evolve, I hope to see solutions that democratize access and bridge the gap between technical complexity and user-friendly implementation.
The academic community offers diverse perspectives on AI agents. Proponents highlight their potential to enhance productivity and innovation. For instance, Azevedo et al. (2024) note that Automated Machine Learning (AutoML) tools can streamline core steps of ML workflows, promoting accessibility and empowering both novice and experienced data scientists. Additionally, a recent video titled “The Future of AI Agents: Revolutionizing Industries” explores how AI agents are poised to transform various sectors, emphasizing their role in automating complex tasks and driving efficiency. Another video that features Elon Musk discusses the transformative potential of AI agents, with Musk stating, “I think it will be able to do anything a human can do, possibly within the next year or two.”
Conversely, some scholars express caution. BaHammam et al. (2023) discuss the ethical challenges associated with AI in scientific writing, emphasizing the need for robust guidelines to ensure responsible use. Similarly, Mustafa Suleyman, in his book The Coming Wave, warns of the unprecedented risks that AI and other fast-developing technologies pose to global order. He states, “Over time, then, the implications of these technologies will push humanity to navigate a path between the poles of catastrophe and dystopia.”
I titled this post GPT-Researcher – On the Edge of AI Agent Viability because this experience has left me feel a bit of concern. We could be on the cusp of a remarkable shift in technology for the good of humanity, or we could be facing a shift that really increases the haves and have nots divide. I have so many questions about accessibility and equity of access to the technology. If the technology, firstly, becomes viable through addressing current complexity and usability issues, for those who have reliable internet access and devices, what happens to those who don’t have the infrastructure. Also, will there be a further divide between those who are tech savvy and the lay person? Will all of this create some kind of neo-technological class divide?
For now, my journey with AI agents leaves me with more questions than answers. How can we ensure these tools benefit all of humanity? What role will accessibility play in their adoption? These questions drive my continued exploration of AI and its potential impact on society.
“I think it will be able to do anything a human can do, possibly within the next year or two.”
– Elon Musk
