Chinmaya

@ChinmayaKB

Product Guy, IoT enthusiast

Bengaluru South, India

Joined April 2014

914 Following

97 Followers

53 Posts

ChinmayaKB retweeted

Santosh Arron

@santosh_arron

28 days ago

I have to push back a little here specifically on Mythos, Fable, and the framing around Anthropic's models. All of these, including Anthropic's Claude family, are built on transformer architecture. There's no fundamental architectural breakthroughs here that India is locked out of. The transformer is public knowledge. Sarvam, Krutrim, and others are building on it too. So the question isn't just whether we can access these models it's whether we should keep chasing the same architectural paradigm at all. And my honest answer is: we shouldn't. We should stand on our own and go further. At Blankline, we've been working on something we call Hope a post-transformer research initiative. The core thesis is that transformers compute the wrong probability operation for general intelligence. They estimate P(next token | context). What intelligence actually requires is P(latent structure | observations), followed by search and verifier-driven reasoning. That's a fundamentally different thing. Our Hope-1 architecture discrete-latent program codes, verifier-driven search, a pre-registered seven-rung validation framework has already cleared four of those seven rungs at 0.69M to 3M parameters, exceeding the closest published baseline by approximately 2× on ARC tasks. The research isn't about building a better LLM. It's about asking whether the whole paradigm is wrong. This matters for what you're saying. If you're right that technology is now the ultimate sovereignty question, then copying the transformer race whether from American or Chinese open-source models still leaves India downstream. What changes the equation is a research bet that could shift the cost frontier of general intelligence entirely. That's what we're working toward. I turned down funding that would have made Blankline instantly credible overnight. Those investors wanted US and UK IP transfer as a condition. If I were just building a product, I might have taken it. But the research we've done and what we have internally right now is something I wasn't willing to hand over. The value isn't in the model weights. It's in the architectural thesis and the research direction itself, which we've published openly at https://t.co/SCO8YKcUTD while keeping the implementation proprietary. India's top investment firms I've spoke with wants profits, not pure R&D. That's a real problem and I've run into it directly. The SF ecosystem understands how transformative foundational research can become. We haven't built that culture here yet but that's exactly the culture India needs if we want to be a source of frontier intelligence, not just a consumer of it. The goal isn't to make intelligence available only to Indians. It's to make it available to everyone and that only happens if someone builds the thing from first principles rather than licensing access to someone else's architecture.

679

108

161

73K

Chinmaya @ChinmayaKB

8 months ago

@ActusDei Personal finance

Chinmaya @ChinmayaKB

about 1 year ago

@djhrishi @pvsubramanyam Nothing is more important than family !

570

Chinmaya @ChinmayaKB

over 1 year ago

@DriveSmart_IN While the two wheeler had the indicator on, he might have chose to slow down or look at the rear view mirror before taking the turn The car driver should have slowed down nonetheless