Dear Reader,

1. In this newsletter, you are going to learn about the Consensus MCP, which became the backbone of my academic AI. It allows your ChatGPT/Claude/LLM to reliably search for papers and interpret the results. You can have it search for papers 100s of times within minutes and run fine-tuned literature reviews or critical paper-backed reviews of your paper's claims. A video of one use case is included.
2. I am also experimenting with a new AI writing method that works in bulk on my entire discussion section and visualises it in a mind map. I recorded a short video for you.
3. Three articles about academic AI use: AI-screened grant rejections, a new scientific framework that challenged a 20-year-old ecological theory, and the story of a professor who got attacked for sharing a paper on AI.

Let me know what you find useful and interesting,

Ilya

Weekly Tutorial: Using the Consensus MCP and AI agents

An MCP connector is a new technical development that gives AI access to external websites and services. In this article, I introduce the Consensus MCP, which lets an AI agent (ChatGPT or Claude) reliably search for academic papers. MCP connectors are a big part of the recently released AI agents course. Consensus is currently the only MCP option that does not require an active subscription (here is my previous review of the tool).

The key mindset shift to use this technology efficiently is to think in goals, where MCP connectors are tools to help the agent. Too often we want to use the tools ourselves (as this was the pre-AI standard), and this gets in the way of truly autonomous processes, which will likely dominate research in the near future.

The tutorial takes you on a step-by-step exploration of this technology and concludes with two larger use cases. As I am still working on two paper revisions, I have borrowed the main examples from my own work for this article:

1. An agent which finds papers supporting & destroying the claims of my paper
2. An agent to build a detailed lit review matrix consisting of 100s of analysed studies.

The article also includes a video walkthrough of the lit review matrix use case.

As always, I have a discount code if you want to explore the premium version of Consensus. Use this link to get a one-week free trial of Consensus Pro or use the promo code effortlessacademic to get 25% off a 1-year subscription.

An AI Writing Experiment

My work on two paper revisions is moving along nicely and I have moved to the writing phase and am experimenting with a new mindmap-based academic writing workflow.

In my opinion, the best method for writing involves using bullet points with references ("atomic sentences") and having an AI draft initial paragraphs from it (see tutorial or full course). These sentences can first be laid out on a mind map for an overview.

Now AI has become powerful enough to do this method in bulk for the entire section. This has advantages:

1. AI can catch duplicates and logical holes
2. Saves time, as you work on 10+ paragraphs in bulk
3. A visual method that shows the evolution of the writing (avoid going in circles).
4. Incorporation of manuscript-level feedback like "the discussion is too long"

I use the Claudian AI agent inside Obsidian and use my academic writing skill, as well as the official Obsidian canvas skill. Here is a 3-minute walk-through of this approach. (Keep in mind this is totally experimental at this stage, and I am unsure if it will actually end up using it, or whether you should!)

The writing draft is organised in columns:

1. My original text
2. 10w summary (for overview)
3. AI suggestion of what should be done with the paragraph
4. AI suggestion of the changed paragraph.
5. (Not shown) Second draft of my text
6. (Not shown) Second round of AI suggestions
7. ...

What do you think? Feel free to reply or comment in the EA community if you're a member.

Three articles from outside EA

​Neutral biodiversity (made with Claude) • Video​

An ecologist and a physicist set out to challenge one of the three major theories on biodiversity: The Neutral Theory. The remarkable part is less their result, even though it is highly relevant to ecologists, but the fact that they used an AI-based framework (BootLoops) to develop this model. Anthropic calls it "claude shaped science" and suggests that by picking the right problems, AI can greatly accelerate science, but not all problems are suitable.

🌲 My take: While my research is far less impressive than what these scientists are doing, I have experienced the exact same tendency. AI is not just a better way to Google, but consistently comes up with new ideas, implements them and leaves interpretation to me. I think any researchers with some quantitative element in their work will benefit from adapting this technology.

​U.K. funder uses 'AI triage' to reject grant proposals

CRANE, a publicly funded U.K. network that supports research on cybersecurity, used AI to triage grant applications. While CRANE claims the AI maintains confidentiality, the process itself was not disclosed to applicants and only AI-generated scores were communicated, causing widespread frustration for applicants.

🌲 My take: With the peer review system overloading with AI-generated content, I think it is inevitable that AI will be used in any, if not most, evaluation processes. However, I think a system like this could be disclosed and made available ahead of time to help improve the quality of applications, rather than used in an intransparent way behind the scenes.

​The AI Discourse in Academia is Toxic. It Doesn’t have to be.​

This is the story of how Ran Blekhman, a professor at the University of Chicago, shared a short opinion post on the Paper2Agent paper and experienced "a toxic, knee-jerk hostility to AI in which the specific tool, application, or question all disappear into a general presumption that the whole category is terrible". (Paper2Agent is a system proposed recently in Nature, which transforms papers into interactive AI agents including the papers' code, content and data.)

🌲 My take: I experienced toxicity around AI first-hand and was almost expelled from university for publishing an AI-based content drafting workflow. The hook of the post was taken out of context and amplified, just as Ran described in his post. Every new paradigm shift initally leads to polarisation, the problem might be that social media algorithms amplify simplistic, and emotional comments, thus precluding the actual debate, worth having. Here is a video for my story, btw.

Want to leave Effortless Academic forever: Unsubscribe from everything. I am sorry to see you go, and farewell.