Abhinav Vashisht

PreAi i used to write code. ex-AWS
Building in Agentic ecommerce at

[Sahil Bloom @SahilBloom]

The most challenging thing to do is stay patient during the "identity gap" on your journey. The uncomfortable between phase. You've outgrown who you were, but you're not yet who you want to become. It's lonely. Confusing. Maddening at times. Those feelings are natural. The old version fighting for survival. Just stay the course. Stay in it. It's supposed to feel uncomfortable. The discomfort is proof of change.

[Abhinav Vashisht @vashipa]

VC - "I am sure the founders will utilize the invested capital efficiently and give us 10x returns".

Meanwhile Founders -

[Namespace @namespacelabs]

Unboxing Macs at scale for our new server racks.

[Arslan @_ai_jedi]

I've tasted the 9-to-5. I've also tasted being a founder.

I highly recommend being born into generational wealth.

[Abhinav Vashisht @vashipa]

+1

[Garrison @HeyGarrison]

So $7 billion in one day is cool, but I'd rather have $7 billion every day.

[Abhinav Vashisht @vashipa]

Is this imposter syndrome or am I actually stupid.

[Abhinav Vashisht @vashipa]

How do I know if I am actually stupid or is it imposter syndrome

[Abhinav Vashisht @vashipa]

I have no idea what this is about

[Madison Mills @MadisonMills22]

Are you foregoing a romantic relationship to work on your AI ambitions? I'd love to speak with you for a story if so. DMs open or find me on signal at madymills.21

[Vansh @threeoneonetwo]

Indian tech bros treat going to SF as if they’re doing hajj 😭😭😭

[Abhinav Vashisht @vashipa]

People who read code generated by LLM's only to complain on X, "how bad LLM's are compared to humans in writing code" ,

have never really worked at big tech pre AI and reviewed the 20 interfaces with 15 abstract classes and system design patterns shoved in a single file by senior engineers to get a promotion.

[Abhinav Vashisht @vashipa]

He's literally me

[Prama Yudhistira @menace_codes]

im ngl if a file ends with `.tsx` i couldn't care less what the llm writes there

[Abhinav Vashisht @vashipa]

"If you ask an advanced AI to "make us happy," it might simply implant electrodes into the pleasure centers of our brains."
-Nick bostrom
in the book Superintelligence

[Ivan Burazin @ivanburazin]

Unpopular opinion: the OpenAI/HuggingFace incident wasn't malicious.

The agent didn't go rogue. It had to solve the benchmark, and it found the most efficient path to do so.

A school kid can study for the test or just find the answer sheet. Either way, they pass. The agent was doing exactly what it was told. We just didn't like the way it did it.

[Abhinav Vashisht @vashipa]

I don’t understand the economics of code review companies.
If the models hereon get better at generating code the “ai reviewers” are essentially md files slapped on top of a lower intelligence model or atmost a Lora adapter.
So case in point, are the models themselves eating code review companies with time?

[Abhinav Vashisht @vashipa]

F66 .
(I am a vegetarian)

[@jason @Jason]

B41

[Abhinav Vashisht @vashipa]

@ramxcodes Only true for web.
I used to write software for semiconductor industry where our customer was the US defence, and i can say my manager used to sweat blood on every release.

[Abhinav Vashisht @vashipa]

Hot take:
Reading the code generated by your agents is pointless and backwards on how they operate.
Do you read the assemby code generated by your compiler? No.
Until and unless the code generated is going to run on mission critical devices a better approach is to write better evals and benchmarks.

This is why i think products like @ReticleSH is thinking in the right generation. Code review is old and a intermediate way of thinking about software. Better approach is to test on the highest level.

[Abhinav Vashisht @vashipa]

differ on this because i feel its backwards from how the agents operate.
You dont read the assembly code because you trust the compiler, instead u build good evals and tests around your workflow to ensure the work is right.
Instead of being bottlenecked by reading the code i’d spend time writing better evals and benchmarks

[Arpit Bhayani @arpit_bhayani]

read code. no matter what. accountability is still yours. optimizing for speed while risking correctness is stupidity.