A personal research journal by Indrodip Ghosh. Long, careful essays on how artificial intelligence really works and where it is taking us, written so anyone can follow, and precise enough for anyone who cannot.
An original study, roughly a year and about $10,000 in the making, on a question I could no longer ignore: how persuasive has a leading frontier model become, and what happens to human judgment when a machine that can change your mind can also do it a million times at once, personalised to each person? Written in plain words from a behavioural-science point of view, grounded in the converging published evidence, and honest about where the real danger, and the real promise, sit.
A story-first investigation into a small, strange observation: the same assistant seems to answer more warmly in Hindi and more bluntly in English. We follow it down through sociolinguistics, honorifics, English-heavy alignment data, and the research on prompt politeness, to ask what is really happening when the language you choose quietly changes the machine you get.
A full findings report synthesizing five decades of research across behavioral economics, services marketing, and psychology to answer one question: what actually governs the decision to buy a coach, consultant, or expert service? Covers the credence-good problem, dual-process decision-making, loss aversion, reference prices, the trust model, social proof, signaling, commitment and pre-payment, choice architecture, and the evidence on whether these services work, closing with an integrated model of the purchase decision.
A technical account of the brain–AI relationship for researchers: representational alignment metrics (RSA, CKA, encoding models), the biological credit-assignment problem and backpropagation approximations (feedback alignment, target propagation, predictive coding, equilibrium propagation), population geometry and manifold capacity, the convergence hypothesis, and the mathematics of the neural read/write interface. Equations included; no simplification.
A deep, first-principles map of the problem, from the neuroscience of the conscious brain and the mathematics of integrated information to a concrete architectural blueprint and a year-by-year roadmap to 2035. Written for the curious layperson and the working researcher at once.
A plain look at what I am building. Vega is our own AI, and Vega 2.0 turns a funnel from a thing you set up once into a thing that runs and fixes itself. It learns your buyer, builds the offer and pages and emails, launches and tests them, and talks to every lead one to one, at scale. Here is how it works, why competent AI is already enough to change marketing, and what it leaves for the human to do.
My working map of where AI is actually heading, written for someone with no technical background. The big shift is from software that answers to software that acts: agents, memory, the quiet collapse of the app interface, models that understand the physical world, and what all of it means for an ordinary working day. Honest about what I believe, and about what nobody yet knows.
A white paper from my own behavioural research, looking at more than 100 American adults aged 30 to 60, on a question that started to worry me: does leaning on AI for everyday thinking quietly weaken the thinking itself? Written in plain words from a cognitive-science and neuroscience point of view, grounded in the converging published evidence, and ending with a practical protocol for using AI without losing your edge.
A first-principles look at artificial general intelligence, what the word actually means, what it does to work and society, and then a blunt thesis about one industry sitting directly in the blast radius. The traditional B2B marketing agency priced the labor of making things. When that labor costs almost nothing, the model breaks. Here is how, what survives, and what replaces it.
Rare, and always in depth. Join the researchers, founders, and simply curious who read each new essay the day it drops.
Occasional, in-depth. Unsubscribe anytime.