Weekend Warp

So this week is unique? - it does not start with, and does not have Altman and Musk in it. The latter is busy interviewing ex-Presidents and the former was growing strawberries, I believe.

This week, the top news is from India (!!), with the country’s most promising GenAI startup, Sarvam , unveiling a host of astonishing products, on a budget and funding which would not cover the salary budgets of a Valley outfit. Sarvam introduced new generative AI products which included Sarvam Agents, Sarvam Models, and multiple other products, with some amazing partnerships with the likes of Yotta Data Services Private Limited , 英伟达 , Microsoft Azure , 谷歌 Cloud Platform (GCP) and @Exotel B by investors such as Lightspeed and Peak XV Partners (formerly Sequoia India), the founders Vivek Raghavan and Pratyush Kumar said its products will be able to perform various tasks and be voice-enabled, supporting 10 Indian languages, namely Hindi, Tamil, Telugu, Malayalam, Punjabi, Odia, Gujarati, Marathi, Kannada and Bengali. This is a big one for the Indian AI ecosystem, and validates Nandan Nilekani’s assertion that India does not need to build its own LLMs, but use clever and thrifty engineering to contextualise and build on top of global ones.

More here

If Sarvam had a big announcement, how could Bhavish Aggarwal of Ola be left behind. On India’s Independence Day on Aug 15, Aggarwal had a slew of eye-popping announcements. First, he announced that Ola will build ‘AI Chips’ named Bodhi-1 and Bodhi-2, supporting over 10 trillion parameters, with Bodhi-2 expected to debut by 2028. Another plan is to have a 1GW Data centre by 2028. The Krutrim AI Chatbot was launched in Android. The ambition to make an AI chip is laudable, though, frankly, I am confused and skeptical on how it will be achieved. Full marks to Ola for its audacity, but lets see what the future brings.

See my comments on it here and here.

Ok, I was being economical with the truth – Elon Musk does get to sneak in. In the middle of interviewing ex-Presidents and stumping for them on X , Musk also managed to launch Grok 2. It claims to be in the Top 5 models, and users (including me) are astonished at its graphics capabilities, which seem better than even GPT4o. However, I am equally astonished by the fact that Grok does not seem to have any ethical guardrails (perhaps, like the founder himself?) and violates copyrights with no embarrassment. Available to all paid X users, it also uses your posts on X as its training data, unless you explicitly forbid it too. Ethan Mollick , well known 美国宾夕法尼亚大学 - 沃顿商学院 professor and AI author, said that: “There are now five GPT-4 class models: GPT-4o, Claude 3.5, Gemini 1.5, Llama 3.1 and now Grok-2.”

Interesting times ahead

The space is certainly heating up....... at warp speed.