Pavan Katta

Co-founder at Hopper Research

[Pavan Katta @PavanKatta1998]

(new post) : make TTS go brrr

Diffusion models produce some of the most natural sounding TTS available today.

But they are pretty slow.

[Brendan (can/do) @BrendanFoody]

When we were starting @mercor_ai, most of the top investors in Silicon Valley told us that the AI data market would collapse.

They had no understanding of the market, but advised us that we should pivot the business to another industry.

The most dangerous thing a founder can do is be too deferential to investors with less context.

Nvidia, OpenAI, Anthropic, SpaceX, Meta, and most of the legendary companies started as deeply contrarian.

If any of the founders had listened to investor consensus, those businesses would not exist.

[Anjney Midha @AnjneyMidha]

ppl are increasingly asking for high quality empirical data on open vs closed frontier model usage

a fairly rigorous paper on this topic is our recent @OpenRouter 100 trillion token study, led by @MaikaThoughts @alexatallah @cclark and team

it is up on arxiv

[Pavan Katta @PavanKatta1998]

they are gonna release voice models soon

[jacob @jsnnsa]

anthropic has ELITE focus

no image models. no video models. no voice models. no world models. three apps. one claude.

this is why they will win. claude is now the best “image model”

[Ben Dicken @BenjDicken]

@jarredsumner We've all been there

[Tianxing Chen @MarioChan2002]

We evaluated 30+ frontier embodied AI models.
The result is clear: current generalist robot policies are still far from robust real-world manipulation.
This is why we built RoboDojo.

[Pavan Katta @PavanKatta1998]

go work with umesh!

[Umesh Khanna 🇨🇦🇺🇸 @forwarddeploy]

🚀 Loving Grok 4.5?

Want to build what comes next at @SpaceXAI?

We’re looking for exceptional engineers who:

1. Are founders or founding engineers who’ve shipped products people love

2. Have trained models and turned them into something people actually use

3. Love building tools that make life dramatically better for engineers and model builders

If that sounds like you, drop your details below. We’d love to chat!

[Malika Aubakirova @MaikaThoughts]

I'm starting 𝐆𝐫𝐚𝐝𝐢𝐞𝐧𝐭 𝐀𝐬𝐜𝐞𝐧𝐭 - a hiking series for AI builders in the SF Bay Area.

A lot of my favorite conversations with founders and operators happen over dinners or happy hours but some of the best ones lately have happened on the move. I wanted to add another format to the mix. Every other month (or so), I'll get a small group together to conquer a hike.

[Pavan Katta @PavanKatta1998]

@MaikaThoughts love it!
gradient

[Pavan Katta @PavanKatta1998]

Qwen3.6-35B-A3B is very underrated

[fal @fal]

First part of our series on how we built one of the fastest image generation systems, and we're kicking off with running LLMs (for prompt expansion) at >1000 tok/s by leveraging a DSpark adapter we trained from scratch!

https://t.co/qGFugOwCPo

[Alex @AlexanderTw33ts]

POV: anthropic team allowing the permanent underclass to have 5 more days of sota intelligence

[Jesse Zhang @thejessezhang]

https://t.co/N2gSVInZVT

[Noémie @FedericoNoemie]

"I can't hang out tonight, we only have two fable days left"

[Robert Yang @GuangyuRobert]

Human brain is a neural net

AI can dramatically improve the training data, just like books and internet did

But it won't fix our algorithm & architecture

And our algorithm is pretty horrible for learning new languages (as adults)

[Aashay Sanghvi @aashaysanghvi_]

OH: “For my next thing, I’d love to be a third, silent co-founder”

[Karina @karinanguyen]

Post-training is beginning to see the light of economic viability, which 6 months ago many VCs didn’t believe in it.

When a model that is 5× cheaper than Opus 4.8 and can beat on PostTrainBench, the economics start to change.

The marginal cost of shaping intelligence is falling, and many enterprise cos start to realize that, we just need to teach them better. That means more businesses can own models trained on their own data, tuned to their own judgment, and improved inside their own feedback loops.

[Pavan Katta @PavanKatta1998]

@nickcammarata stop bragging

[Pavan Katta @PavanKatta1998]

probably the most remarkable post training results I've seen. this should be more famous

[Mira Murati @miramurati]

Bridgewater used their unique financial knowledge and partnered with us on @tinkerapi to fine-tune a model that helps their analysts focus on what's important. Experts improving AI that empowers experts.
https://t.co/6RJITMG2BJ

[Rhys @RhysSullivan]

fable coming back to my codebase after 19 days of opus

[Pavan Katta @PavanKatta1998]

@blip_tm i'm guessing the market is supply bound?

[Jai Bhatia @jaixbhatia]

granola is refreshing because you can tell the product has focus

it feels like everyone it putting out the same "1000+ connector" ai agent, and something about it feels like a race to the bottom