Kiran Kannar

Looking back on the last year of shipping ML systems, I wrote a short reflection on what actually drove iteration speed. Despite coding agents and faster V0s, the fundamentals haven’t changed much: evaluation discipline, realistic data, and tight feedback loops still matter most. My takeaway: Speed is a consequence. Measurement is the work. kirankannar.github.io/nlp/2026/01/18…

Bottle this feeling 🍾

This dogged mentality is so inspiring. So proud to be a Gooner! Congrats @Arsenal

North London forever 🎇

The Arsenal. Your Premier League champions.

This belongs to all of us.

The nature of building is changing in ways that are hard to ignore. We are all architects now. Congrats on the promotion. New blog post: kirankannar.github.io/journal/2026/0…

A corner of Arsenal, wherever you are ⚡ Download the new digital home for Gooners now📱

"Answer thrashing" is an interesting phenomenon & it seems like an example of the scientific value of faithful CoT. (h/t lesswrong.com/posts/btAn3hyd…)

Activation oracles are a technique where a model is finetuned to answer natural language questions about another model's activations. We applied them to a bunch of safety-relevant tasks and got little use out of them, and found them very hard to evaluate.

New blog post: infinitus.ai/blog/using-con… We explore using lightweight linear probes as second-stage models to accompany LLM judges. These act as semantic filters that improve precision while maintaining high recall constraints for safety problems like adverse event detection. This was a fun research project to work on!

@fchollet The people who say SaaS is dead also seem to be the people who don't understand why companies like Salesforce exist. It's not about the code; it's about the data that flows through these platforms and the long-term journeys of customers & their users on them.

The maximalist form of my thesis is basically this: SaaS is not about code, it is about solving a problem customers have and selling them the solution. Services + sales. If the cost of code goes to *zero*, SaaS will *not* go away. It will *benefit*, since code is a cost center.

Unfortunately, I can no longer read a "it's not this, it's that" construction without immediately raising my AI hackles. Just total rhetorical style death.

A rough example is modifying a circuit within the NN to encode desired semantics. Even though ultimately all of it goes through backprop, we can aim to constrain gradients to be aligned directionally with desired internal "structure", rather than letting semantics emerge arbitrarily.

Lots to ponder over in goodfire.ai/blog/intention… by @banburismus_ around intentional design of internal representations of neural networks.

Wrote a new blog post "5,000 Lines of Trust: Engineering Agency in the Age of Agents" kirankannar.github.io/journal/2026/0…

@myra_deng @GoodfireAI Congrats!

We’re (re)building AI the way we think it should be built @GoodfireAI And hiring a lot! Join us

Reading a book is a great way to reactivate your mind.

Love this new era of interpretability work for scientific advancement! 🚀

What interests me in 2026: kirankannar.github.io/journal/2026/0…

Alrighty. What's going on with the spider charts in the background? Is ChatGPT's spidey sense tingling when it acts as my thinking partner?