Dear Reader,

Obsidian had a few updates this month that I think will delight anyone already using the tool, especially academics. Here's what else this week's newsletter is about

1. EA Tutorial: Obsidian 1.14 update on coloured highlights and academic bases.
2. Free recording from my Litmaps webinar last week: The detective mindset for literature reviews.
3. Three stories across the world: AI bias, conferences overrun by AI papers and the troubles of synthesising ideas in science.

Let me know what you find useful and interesting,

Ilya

Weekly Tutorial: Obsidian 1.14 Update for Academic Use

Alongside many small improvements, two main updates stand out: Kanban Bases and Coloured Highlights. The former makes Obsidian much more usable for task management, and the latter is a much-anticipated feature, especially if you are annotating texts in Obsidian.

Kanban Bases: If you have never heard of bases, think of them like Microsoft Excel. Each row is a note or file in your vault, and each column is a property of that file (like creation date, or "Type"). Just like in Excel, you can filter and group by specific columns.

But Bases go beyond Excel and allow you to display the results not only as a table, but as a list, gallery, and now as a Kanban board as well. In previous posts, I wrote about how incredibly useful Bases are for organising your ideas, finding hidden connections, and organising meetings.

With the new update, they're also useful for project/task management. They are foundational for large-scale information management and are covered in the Information Management Course.

Coloured Highlights: An often-requested feature is to have multiple colours when highlighting text in Obsidian. However, because Obsidian is text-only, this isn't easy. Markdown supports highlighting only through ==highlight==, not different colours. Plugins solved that by using HTML tags, but they "pollute" your text-only notes with invisible code to preserve the appearance, and nudge you toward needing software to read your notes in the future. Just like you can't easily read a website as text, as it's mostly HTML tags defining the appearance rather than information. The new update changes that. I love this new feature. With some help from my AI agent, I created an overview of changes to a manuscript. (Finally, replacing MS Word's "track changes", which comes to a crawl at the size of this document.)

Take a look at how it all works in the recent post.

Free Literature Review Webinar Recording

A few weeks ago I gave a webinar on how to use Litmaps, Zotero and Obsidian for a Literature Review that does not feel like having to read through an entire library. In the Lit Review Course, I call this approach the "Detective Mindset". It works by moving step-by-step, updating your notes, and frequently changing the active question you are trying to answer. Litmaps is a key tool in this journey, making sure you don't miss any papers. The result is a fun, effortless, and effective literature review process. Check it out if you have 45 minutes to spend:

ps: If you haven't yet, please join my YouTube Channel. I am working on new videos here as well!

Three articles from outside EA

​AI Bias in one graph • X post​

The graph speaks for itself: asked to pick a number between 1 and 10, AI picks 7 more often, while humans do this, too, but to a far lesser degree (lucky number?). While I am not sure how this graph was generated, or how truthful it is, it exemplifies a fundamental attitude we need to take towards AI.

🌲 What this means: The point of this tweet is that humans say "seven" slightly more often, which is then picked up and amplified by AI, making it far more often. Gradient descent here refers to the way models are trained: "Reducing its error function by taking the step with the highest reduction, i.e. the gradient", in this case, learning to reply "7". This shows that biases can be amplified when AI models are trained with large societal implications. This reminds me of an intriguing essay arguing that Western values and biases (on which AI is built) might be out of place in other cultures.

​Computer Science Conferences impose submission caps due to AI "Tsunami"​

"Ten years ago, the International Conference on Learning Representations (ICLR), [...], received about 500 submitted papers. When the deadline for the 2027 meeting passed last month, the number had ballooned to more than 40,000". As a result, conferences are putting caps on the number of submissions. The likely reason is the advent of generative AI (the article shows an impressive graph).

🌲 My take: There is a gold-rush feeling at the moment motivated by (1) the publication quantity and citations as academic currency, (2) the easy access to generative AI, making it easy to create and (3) the lack of quality assessment automation (for good reasons!). Collectively, we see a similar development on social media: A rise in new content, as people try to figure out how the new paradigm works. I don't like to use the term "slop", as some of this content/papers aren't bad, and some AI-generated content is genuinely a new "medium". However, I think this points to a need for some form of automated content assessment to encourage quality over quantity.

​How to assemble science​

In this essay, Helen Pearson argues that science's real bottleneck is often not a shortage of knowledge, but a failure to synthesise what already exists. Systematic reviews and meta-analyses reveal patterns invisible in any single study. For example, Sudden Infant Death (SIDS) risk from prone sleeping was synthesizable by 1970 but went unacted-on for two more decades, an estimated 50,000 preventable infant deaths later. Climate Science is another such example.

🌲 My take: I think this is precisely where academic tools and AI can be so brilliant, as they can amplify the sometimes dull task of assembling 100s of studies necessary for meta-reviews. However, traditional AI tools are often unsuitable here, as they are not reproducible and most scientists lack the skills to build custom tools. Tools like Nested Knowledge are attempting to fix this, but are not well known in the community (and are expensive). There is an innovation gap here.

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