𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻

You don't know enough until you can do everything you can think of.

Tweets are notes to self.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

@opencode We learnt it the first time:

Permenent = Until the provider increases the prices again 🫠

(Secretly hoping I'm wrong 🥲)

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

@CommandCodeAI @rails recently said, and I quote "DeepSeek 4.1 Flash, figured out it was being benchmarked and tried to hack its way to a better score."

What are the chances the same is happening here?

If anything, it's a complement to DS- it's so good, it's almost suspicious lol.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

But it does not always translate to something meaningful. To do that, one has to be intentional, strategic and meticulous.

Looking back, most of what I built lacked character that could make me proud. The right thing would've been a better foresight and vision for the projects.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

Being relentless has hurt me in the past 10 days. When I went on a vibe coding spree, I felt like I was addicted, negatively. I lost focus, memory, will power, and energy. Sometimes losing sleep isn't the sign of hustle. Watching things happen quickly is just a pleasant feeling.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

I hope to see labs now focus on tiny, ultrafast classification models that can deliver sub-10ms inference and much lower compute, and operational complexity of ML systems (and could be locally hosted on a low-end hardware).

🗣️ Not everything needs to be generative.

5/5

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

LLMs had become the fast food of software: convenient, addictive- you couldn't have enough of them, made you bloated and slow.

One of the unhealthy effects of the LLM boom is that it made generative models feel like a silver bullet to every problem.

1/5

[Ashutosh Maheshwari @asmah2107]

Jev is the first time AI feels designed for backends.

Remember the feeling when GPT-3 dropped and you realized the rules of software had changed forever?

I had that exact same feeling messing with Jev yesterday.

For 3 years, we’ve been forcing generative, autoregressive models into places they probably didn’t belong.

The mental model with Jev feels completely different:

LLMs → slow, deliberative, chatty, heavy
Jev → sub-80ms, parallel, calibrated state

Instead of asking an LLM to generate its way through API plumbing, you get something that feels purpose-built for backend systems.

Maybe the interesting shift isn’t making LLMs faster.

It’s building AI that doesn’t need to behave like an LLM in the first place.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

I still love LLMs, but they had been cursed by our own tendency to overfit all aspects of our interests with anything that remotely works.

Thanks @typesafeai for lifting that curse and getting us to see that light.

4/5

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

Congrats @typesafeai for breakign the curse of LLM.

Hopefully labs will now focus on tiny, ultrafast classification models that can support sub-10ms inference and dramatically lower the compute, and operational complexity of ML systems.

🗣️ Not everything needs to be generative.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

If the output is one of 20 labels, you probably don't want a generative model.

The right model isn't necessarily the biggest or most general one.

It's the smallest model that solves the problem well.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

Maintaining equanimity while making sacrifices to get past hindrances leads to peace.
Peace leads to further focus.
All of this leads to further enrichment of character.

2/2

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

All pursuits are pointless without Adhisthaan.
Adhisthaan leads the first hammer-strike on the sculpture of character.
Character leads to clarity.
Clarity leads to simplicity.
Simplicity cultivates focus.
Focus passes by hindrances.
Observing the hindrances leads to insight.
1/2

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

@CommandCodeAI Desktop & @zeddotdev Delta (public beta), 2 releases was looking forward to dropped within 12 hours. 👏

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

@Amul_Coop @FDA_MAHARASHTRA @fssaiindia

I ordered 4 cartoons of Amul Tazaa milk from
@letsblinkit. 3 have the same packaging while 1 is much taller, with noticably different cardboard quality.

Could you confirm if this is expected?

How to verify its authenticity?

Thank you.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

@MrAhmadAwais @ziwenxu_ Understand your frustration as a founder. People have no idea how unit economics get messed up when subsidized services are not used in the spirit of the core offering. The $1 plan got me to try CMD and now I'm thinking of going for GOAT. Hope it doesn't come to discontinuing Go.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

@ixigotrains Appreciate it.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

I just missed an urgent Tatkal ticket because I trusted @ixigo over my own judgment. 😡 I knew the window opens at 11 AM, but their app UI explicitly told me bookings start at 11:30 AM. Trusting a major aggregator cost me my seat, everything was completely gone.

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

@ixigotrains @ixigo @RailMinIndia IRCTC's 30m blackout is a good regulation, but @ixigo your text is misleading. Saying "Bookings open at 11:30" (screenshot) tricks desperate passengers into logging in 30m late. By then, seats are gone. Show BOTH times- people pay extra under urgent needs. @RailMinIndia

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

@ixigo @RailMinIndia And the misdirection continues...

[𝙰𝙱𝙷𝙸𝚂𝙷Ξ𝙺 𝙿𝙰𝚃𝙸𝙻 @maverikkano]

Honestly makes me wonder if this "glitch" is intentional. Misdirecting regular passengers to log in 30 mins late just leaves the field clear for bots and unauthorized agents to sweep up seats on IRCTC. Absolute rubbish experience. Fix your UI @ixigo! @RailMinIndia