Prem Viswanathan

Building @swift_cx. Adjunct @ CMU. Prev @aws

[Prem Viswanathan @prempv]

Insane growth.
The product is even better.

[Simon Eskildsen @Sirupsen]

turbopuffer crossed $100M run-rate in March. 19mo after $1M. Profitable & <$1M raised.

Cursor・Anthropic・Notion・Cognition・Harvey・Bridgewater・Ramp・Linear・Legora・Superhuman・Atlassian・Granola

We’d be nowhere without them. We work like hell to exceed their expectations.

[Prem Viswanathan @prempv]

@nikunj True, very competitive model.

But the types of hallucinations I've encountered at times are a bit scary for critical production use.

E.g: it used Hindi words randomly when drafting an email. This is from API - Hindi was nowhere in the scene.

[Prem Viswanathan @prempv]

This was one of the things Claude got right early on, when it was generally subpar.

https://t.co/EznQxbt6xb

[adam @theCTO]

hey @sama can we normalize models just saying "i dont know" ?

eliminates 99% of hallucinations

[Prem Viswanathan @prempv]

@nateharada 😂 lol

[Prem Viswanathan @prempv]

Let's go 🚀

[SpaceX @SpaceX]

SpaceXAI and @cursor_ai are now working closely together to create the world’s best coding and knowledge work AI.

The combination of Cursor’s leading product and distribution to expert software engineers with SpaceX’s million H100 equivalent Colossus training supercomputer will allow us to build the world’s most useful models.

Cursor has also given SpaceX the right to acquire Cursor later this year for $60 billion or pay $10 billion for our work together.

[Prem Viswanathan @prempv]

@HarryStebbings If you're paying in API, the cursor harness in max mode is better with Claude models.

[Prem Viswanathan @prempv]

I like this!!

[Brendan Falk @BrendanFalk]

I believe we've found the best AI-native coding interview

We call it the “Composer 1 interview”

Candidates get 1 hour to build a real, medium-sized project live

The only constraint: they have to use Cursor’s Composer 1 model

[Prem Viswanathan @prempv]

@nikunj create your skills*

[Prem Viswanathan @prempv]

@nikunj Use this plug-in to create / update your plugins.

It's helpful to figure out all the possibilities with skills. https://t.co/IS3RnKhS5q

[Prem Viswanathan @prempv]

Plot twist @openclaw changed the password.

[Lenny Rachitsky @lennysan]

I lost track of my Mac Mini password where I have @openclaw running, AMA

[Prem Viswanathan @prempv]

This is incredible! Very competitive from Grok team.

[xAI @xai]

Today, we're excited to launch the Grok Voice Agent API, empowering developers to build voice agents that speak dozens of languages, call tools, and search realtime data.

https://t.co/7c7SLYzvum

[Prem Viswanathan @prempv]

Claude Opus 4.5 is definitely having some hiccups.

Similar quality issues with @WisprFlow this evening.

AI quality degradation is the new "stackoverflow / aws is down" moment

[Prem Viswanathan @prempv]

Incredible!! Congratulations.

[emozilla @theemozilla]

@NousResearch @hillclimbai two decade long revenge arc on my 2.9 real analysis gpa finally complete

[Graham Neubig @gneubig]

We're hiring! We have positions open for Members of the Technical Staff for Agent R&D and many other positions.

Think of the best researcher or engineer you know, don't you want them building in the open?

Listings below! https://t.co/7x6QI2Wuj6

[Prem Viswanathan @prempv]

Can’t wait!!

[Nick Miller @nickwm]

Working at @cursor_ai feels like riding a rocket ship. The momentum’s been accelerating all year and what’s coming next week will take it to a new orbit.

[Prem Viswanathan @prempv]

Compression's always relative: today's model capacities were sci-fi 10 yrs ago.

For coding, we thought we would need 100M context. But smaller context with many parallel exploration and convergence is winning.

I expect the same here to solve the end goal - how does one go about reading McKinsey slides.

[Prem Viswanathan @prempv]

@_cartick @dileeplearning We humans read text visually isn't it.

With sufficient resolution surely this isn't a problem.

Vision and audio as the two universal input modalities makes a ton of sense IMO.

[Prem Viswanathan @prempv]

@arafatkatze @andrew_melby @cline @AmpCode @Cursor When you refer to RAG, you are essentially talking about pure vector search, which is indeed quite problematic. But having it as an optional tool should be fine.

Your concern about the overhead of vector search relative to the lift it offers is a very valid and quite Underrated.

[Prem Viswanathan @prempv]

@gneubig Emails with resend? Is that replying to Gmail with resend?

[Prem Viswanathan @prempv]

This would be a great team to be part of.

Retrieval & Context for AI is everything!

[Gabe Pereyra @gabepereyra]

We are hiring for Harvey's Retrieval and Data team to expand Harvey to over 1,000 data sources in the 50+ countries we operate in.

Our clients use Harvey to do research over legal, financial, tax, and regulatory data around the world.

Thread on search + data at Harvey: