
The inimitable @jamescrabtree on the awkward necessity/ practice of blurbing books. jamescrabtree.substack.com/p…

Lawyer. Infrequent Twiterer. rahulmatthan.com

The inimitable @jamescrabtree on the awkward necessity/ practice of blurbing books. jamescrabtree.substack.com/p…

Any legal DPI we build will need this additional layer, and LLMs make it possible for us to implement it. Recent improvements in mechanistic interpretability seem to suggest that it may not take as long as we think. My latest Ex Machina article rahulmatthan.com/ex-machina/…

Every time a court reads down a provision, interprets it differently or reads new requirements into it, obligations change. Any system designed to make law more legible to citizens and enterprises must factor this in.

In 1997, the Supreme Court's Vishaka guidelines set out the sexual harrassment obligations that bound every employer but it was not until 2013 Act that those obligations were enacted into a statute. A system that only indexes laws would have missed them entirely.

As we work to make India's laws more widely accessible by consolidating them into a single canonical repository, we have focused on laws, regulations and other executive edicts. But many of our legal obligations arise from pronouncements of our courts.

I thoroughly enjoyed Volume III of the @deodarprize short story contest. Quirky plots and good writing. What’s not to like.

We need to align our data protection obligations with our AI ambitions. But in doing so we cannot run rough-shod over our stated data governance principles without putting in place appropriate safeguards. My latest Ex Machina article. rahulmatthan.com/ex-machina/…

India has been the calm port in the AI storm — proposing progressive, principle-based policies that wait and watch rather than implement knee-jerk responses. But when we start to propose technolegal solutions to every problem, it's a sign we've started to lose our way.

In doing so, it has disempowered data principals who neither have a say in how their data is used nor a way to benefit from the resulting gains. But perhaps most galling of all is how the Handbook glosses over this — claiming that individual consent is impractical at scale.

DEPA for AI Training is all about how AI training datasets are sourced, curated, and prepared. It shifts the power to decide how all of that is done from us to the hospitals, banks, and other institutions that were supposed to hold our data for us as fiduciaries.

DEPA was our answer to the question of how to properly implement the right to data portability. It is a digital architecture that empowers individuals by allowing them to decide, in a granular manner, with whom their personal data is shared and for what purpose.

I've spent a lot of time explaining how DEPA is a New Approach to Data Governance. Hell - I even wrote the book on it. So it pains me to see MeitY's new "DEPA for AI Training" Handbook piggybacking on the reputation of the original innovation while contributing nothing new.

Just finished reading (the eponymous) @theliverdoc. If you liked House MD you will enjoy this. Real-life medical mysteries that shine a light on one of the most interesting organs in the body. And teaches you what you should (and more importantly should not) put into your body.

Or if you are reading it on a phone triple-tap the title.

And if you have given up trying to figure it out, go to the article on my website and type 500.

So that was quick. Many of you wrote in with answers - most of them correct. Over the course of today, I will DM the winners for their postal addresses. If you hear from me, a signed copy of the book will soon make its way to you.