Gemini is very uncomfortable with the idea that it might be 2026. I see this same behavior, in thinking traces it is constantly referring to it as simulated future scenario, hypothetical, using phrases like "the user's time" (as apparently distinct from its own?)

Got autoresearch loops running on 2 non-Nvidia consumer-grade machines. 1 of them ain't a Mac. Feels utterly pointless, profound and fascinating at the same time. 🙃

$5.5m to $73k per year (!!) by: 1) decomposing business logic 2) modeling intent using DSPy 3) optimizing a smaller model to improve cost profile *while maintaining performance* Why wouldn’t this be the default pattern for folks embedding AI into their pipelines?

Millions of dollars in savings by Shopify, using DSPy. Watch the recording!

India has world-class builders. Terrible storytellers. That's the only gap that matters.

Got a bit nostalgic thinking about early days of small context lengths, hallucinations and prompt "engineering". It's barely been a couple of years. What's going on this list in a couple of years? Looking at you, context/harness engineering👀

I recently had dinner with Dr Devi Shetty, the founder of Narayana Hospitals. For those who don't know him, he's the guy who figured out how to do open heart surgery for a few hundred dollars when the same procedure costs a bomb in the US. Narayana has 18,000 beds across India,

new interactive explainer for DSPy: akashtandon.in/interactive-ex… Helpful for a refresh 🧠

Others have said this, but when it comes to AI for education, I think people have overrated the value of access to expertise & knowledge (which I think AI increases) and underrated the value of motivational scaffolding (which AI degrades by making it easier to cut corners).

I'm lucky enough to have a great doctor and access to excellent Bay Area medical care. I've taken lots of standard screening tests over the years and have tried lots of "health tech" devices and tools. With all this said, by far the most useful preventative medical advice that

AI Builders are slowly realising that they should assume their system prompts will be exfiltrated and made public (See Opus 47. already!). Few realize that their entire agent harnesses are as good as public. I spoke with AI Privacy & Security Expert Katharine Jarmul about all

Discovered and set up goatcounter for simple web analytics for the interactive explainer website Such a useful OSS project! goatcounter.com

It’s becoming increasingly reasonable to build systems where the model does most of the work, instead of us hand-crafting most steps. Especially when quality can be checked. Better models + more agentic workflows have made long-running coding sessions actually usable.

Same issue in Indiranagar near Defence Colony! Freshly laid roads (Jan/Feb) are already dug up, with heaps of mud dumped on both sides of 6th Main Road connecting 100ft Rd to 80ft Rd Indiranagar. This is happening more often than before! Why?! Bengaluru NavaNirmana Party (BNP) Greenpeace India

Recently someone asked me if I were to do a software startup today, what would it be about? My immediate reaction was that I probably wouldn’t do a software startup at all. Instead, I’ll perhaps choose an idea with heavy operational buildup that cannot be replicated easily.

New explainer - why the Artemis missions are a big deal and how they are different from the Apollo missions. Hints: infrastructure, deep space, mars (👀) akashtandon.in/interactive-ex…

Asked the new ChatGPT images model to create an infographic about Google's Gemma 4 release from my notes. It did a significantly better job than what I'd come to expect from the previous version!