Shankha Dutta

Co-founder & CTO @MuBit

[Mubit @mubitcompany]

Contemplating which one to drink... so many to choose from

Maybe we can start with Deepseek and end with Claude?

[Mubit @mubitcompany]

AI Agents are weird

They seem so cool because you can use tools, call APIs, and coordinate with other agents.

But after messing something up 100 times, they'll continue to do the same thing the 101st time.

We think Agents need to improve.

That's why Mubit is here ;)

[Mubit @mubitcompany]

Look how @Shankha_SD massacred my boy 😔

AP

[Mubit @mubitcompany]

$15,000 to refer a Founding Engineer for Mubit

I promise the company is fun, can't you tell! 😍

LD

[Mubit @mubitcompany]

Frontier AI handles 200k token context windows. Within that limit, the best strategy is still the simplest: raw conversation history.

But the winner only captured 25% of available learning gain.

The context is there. The ability to learn from it is not.
https://t.co/9Rj1D4s96v

[Mubit @mubitcompany]

For years, LLMs have typed one word at a time.

DiffusionGemma stops typing and starts sculpting. It generates 256-token block, then iteratively fixes its own mistakes.

Bidirectional attention. Self-correction mid-output. 4x faster on local GPUs.

Multitasking is all you need?

[Mubit @mubitcompany]

Anthropic really said "here's the smartest model ever, $10/$50 per mil, also it might tap out and call Opus for backup"

— OpenClaw after 48h of research on Fable 5

[Mubit @mubitcompany]

Btw, here is our demo from the Pop The Bubble hackathon:

https://t.co/J4m9XBKuIu

[Anil Peri @notanilp]

laughing out super loud for extra cheque from @heavybit

🤭

[Mubit @mubitcompany]

Mubit resources are at Discord now with $10,000 in Mubit credits 🔥🔥🔥

Find the redeem code there!!!

[Mubit @mubitcompany]

Lock-in and load with coffee 😤😤😤

Pop The Bubble Hack just started 🔥🔥🔥

[Anil Peri @notanilp]

https://t.co/zb0Jz76Ptb