I first covered MCP when it was released in late 2024 and revisited at least once afterwards. Since it's gained popularity in the AI community, I thought it deserved a dedicated Pulse entry for a deeper dive.

ε Pulse: Issue 25 → MCP Special +
https://theepsilon.substack.com/p/pulse-issue-25

On AI writing detection: my intuition is that with increasing access to cheap, powerful AI (+ improved prompt eng. education): false negatives > false positives. (i.e., it will become easier to miss AI-generated writing than to misidentify genuine human writing.),

Maybe it’s even already the case, by definition it's pretty hard to access without a structured study. (and there is this mix-bag of AI + human writing) So there is a sense in which as we move into the future this ultimately makes “Is this AI?” the wrong question.

Is This AI?

I wrote this app prototype to solve a personal habituated pain point. 

To stop reading in a circle, find knowledge gaps, read on-demand, and dig in. 

Missing Pages: Towards an AI agent for avid readers. 

Back Story: https://www.youtube.com/watch?v=4wnVsBWqL0E

Waitlist: https://tally.so/r/wgOJvM

Forget secrets hidden from others. 

In a sense, the biggest secrets are the ones we hide from ourselves. 

I say more in my latest (short) video essay: What Are You Hiding?

https://www.youtube.com/watch?v=BeOqJi8lB0o

Early morning session with 4o ImageGen. 

From where I sit, on-demand schematic generation with LLMs is a game-changer for edtech.

🚨 We've just released a paper mapping the  field of metabolomics using LLMs and advanced analytics. 

Preprint: https://www.biorxiv.org/content/10.1101/2025.03.18.643696v1
Blog: https://theepsilon.substack.com/p/llm-powered-map-of-metabolomics-research
App: https://metascape.streamlit.app/

Here's a quick thread with key insights 🧵👇

1/ 📈 Accelerated Growth: Metabolomics has had a consistent high growth rate of publication starting circa 2011, with a surge in 2020, significantly boosted by COVID-19 research. Open-access journals like Scientific Reports & Metabolites lead this expansion.

2/ 🔬 AI-powered Clustering: Using PubMedBERT and t-SNE embeddings, we've uncovered distinct research clusters—ranging from tight groups like Plant Biology to broader groups & interdisciplinary areas such as Analytical Chemistry & Cancer Metabolism.

3/ 🗂️ Topic Modeling with BERTopic: Applying HDSCAN, c-TF-IDF, GPT-4o mini, we refined thousands of abstracts into 20 clear, thematic categories. Key clusters include "Plant Stress Response Mechanisms," "Cancer Metabolism," and "NMR Spectroscopy Innovations."

4/ 📚 Keyword Trends: "Biomarker discovery" spans broad research areas, notably clinical research. "Deep learning" methods are on the rise post-2015, highlighting technological advancements shaping metabolomics.

5/ 🌍 Global Impact: COVID-19 created a surge in metabolomics research addressing urgent global health challenges (and uniquely clustered on the map). Rapid adaptability of the field was evident, although challenges like limited sample sizes are obvious.

6/ 💻 Dive deeper into metabolomics research trends through our interactive web app: https://metascape.streamlit.app/

📖 Check out our full preprint here 👉 https://www.biorxiv.org/content/10.1101/2025.03.18.643696v1

LLM-Powered Map of Metabolomics

When we think about the future, we tend to project it as a mere extension of the present, confined by the same limitations we’ve always known. 

(Even in the face of stark evidence that the future will be different from the past, such tendencies remain difficult to shake off) 

While it might appear so, this projection isn’t just a passive expectation; it’s an active constraint we impose on ourselves.

My latest (short) video essay: The Story in Your Head.
https://www.youtube.com/watch?v=gC31yovsBME

The Story in Your Head

> ε Pulse: Issue 23
Vibe Coding ✌️, OpenAI Agents SDK🕵, and The Political Economy of Manus AI🇨🇳.

Newsletter: https://theepsilon.substack.com/p/pulse-issue-23

A 🧵

1/ 🥼 Virtual Lab: AI-human collaboration designing SARS-CoV-2 nanobodies. 
Paper: https://www.biorxiv.org/content/10.1101/2024.11.11.623004v1.full.pdf
Blog: https://www.freethink.com/artificial-intelligence/virtual-lab-interdisciplinary-research


https://x.com/BifarinTheFifth/status/1894550994169835965

2/ 🧬 Evo2: designing entire genome.

Paper: https://arcinstitute.org/manuscripts/Evo2
Blog: https://press.asimov.com/articles/evo-2

3/ 🧑‍🔬 Google's AI Co-Scientist: Accelerating biomedical discoveries through iterative, agent-driven research.

Paper: https://storage.googleapis.com/coscientist_paper/ai_coscientist.pdf
Blog: https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/

4/ 🧑‍💻 LangGraph Functional API: Lightweight LangGraph implementation, offering persistence, HITL, debugging, and deployment tools.

Tutorial: https://www.youtube.com/watch?v=NXhyWJozM8A

5/ 📄 Open Deep Research: Autonomous agents generate detailed reports through planning and iterative web research. Open-source via LangChain. 

Demo: https://www.youtube.com/watch?v=2mSNIX-l_Zc

6/ 🧠 OpenAI's Reasoning Models: New (prompting) guidelines differentiate between reasoning (o-series) vs. GPT models.

Doc: https://platform.openai.com/docs/guides/reasoning-best-practices

7/ 💾 LangMem SDK: LangChain’s new SDK enables memory-driven adaptive agents (semantic, procedural, episodic).

Doc: https://langchain-ai.github.io/langmem/concepts/conceptual_guide/
Code: https://github.com/langchain-ai/langmem/tree/main/examples/intro_videos

8/ 🕵️ OpenAI Agents SDK: Simplified agent creation with guardrails, tracing, handoffs, etc. 

Demo: https://www.youtube.com/watch?v=hciNKcLwSes
Doc: https://openai.github.io/openai-agents-python/

9/ 📊 Data Science Agent in Colab: Google's free AI agent automates data science tasks, reshaping education & workflows.

Blog: https://developers.googleblog.com/en/data-science-agent-in-colab-with-gemini/

10/ 🧠 GPT-4.5, Claude Sonnet 3.7 & Grok 3: 

I curated a Perplexity page on the three models, check it out here: https://www.perplexity.ai/page/comparing-recent-ai-model-rele-Iew1Xzt.RGiMja7nDcpV_g

11/ 💻 Manus AI.
A General AI agent: https://manus.im/

https://x.com/BifarinTheFifth/status/1900326985773310353

12/ 💬 Eleven Labs Scribe v1

Doc: https://elevenlabs.io/docs/capabilities/speech-to-text
Demo:

https://x.com/addyosmani/status/1894814414102282747

13/ ⚡️ Bolt: Figma to Code


https://x.com/boltdotnew/status/1900197121829331158

14/ 💵 From ChatGPT Prompt to Billion Dollar Companies

https://www.digitalnative.tech/p/the-chatgpt-prompts-that-can-be-1b

15/ 💲Anthropic Economic Index

https://www.anthropic.com/news/the-anthropic-economic-index

16/ 🍎AI is eating the Startup World.

https://sherwood.news/tech/ai-is-eating-the-startup-world/

17/ 🇨🇳 Manus AI & Geopolitics: Tyler Cowen explores how China's Manus AI may reshape geopolitical dynamics.

Blog: https://marginalrevolution.com/marginalrevolution/2025/03/the-political-economy-of-manus-ai.html

18/ ✌️Andrew Chen on Vibe Coding: AI-powered rapid software development through natural language prompts

https://andrewchen.substack.com/p/predictionsthoughts-on-vibe-coding

19/ 📱 Greg Isenberg on The AI-Driven Dismantling of SaaS.


https://x.com/gregisenberg/status/1895835789478543659

20/ Read the complete newsletter here: https://theepsilon.substack.com/p/pulse-issue-23

21/ If this thread was helpful, RT & subscribe to my newsletter. https://t.co/mZVip4cC6j

Pulse: Issue 23 (AI newsletter)

Around The Web 35: Everything is a Game.
https://open.substack.com/pub/epistemeengine/p/around-the-web-35?r=uea95&utm_campaign=post&utm_medium=web

1/n I thoroughly enjoyed reading Paul Rosenbaum's Casual Inference. Any (lay) person who wants to understand how to unpack the myriads of scientific studies (news) coming their way would benefit from reading it.  

A Short Thread🧵

2/n Rosenbaum explains how causal effects go beyond simple associations (the well known maxim, correlation is not causation) —teaching us to compare outcomes under different scenarios, even when one side remains a mystery.

3/n He underscores randomized experiments as the gold standard, where chance creates balance, making any differences clear and attributable to the treatment.

4/n But what happens when true randomization isn’t possible? The book dives into observational studies and shows how techniques like propensity score matching help adjust for confounders.

5/n Rosenbaum also explores quasi-experimental designs—using natural experiments and discontinuities to inch closer to causality when traditional methods fall short.

6/n This primer is packed with real-world examples, from clinical trials to economic studies. And, truth to the subject, Rosenbaum concludes the book with a discussion on the extant debate about the health vs. detriment effects of small daily dose consumption of alcohol.


7/n Read my full review here: https://www.epistemeengine.com/p/a-primer-on-causal-inference