Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile
Golden_Iris⚡️

@golden_iris7

building things I'd like this world to have.

ID: 1751885469233942528

calendar_today29-01-2024 08:30:57

563 Tweet

36 Takipçi

289 Takip Edilen

Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

I hate how patriotic I am. The more I learn about India, the less hopeful I become about our "superpower" dream. BUT... then it makes me want to do something about it and contribute towards that dream. It's like a never-ending loop. I become less hopeful and then more motivated

Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

Yk know if job market is changing fast, then just follow a roadmap that changes with it. Sounds simple, tiring to pull off and keep going.

Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

If you think models are too agreeable, a simple way is to just add "answer like the top 1% " or some form of the same thing to get the best result. You'll notice better thinking, too.

Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

Gemini 2.5 pro outperformed gpt-5 in writing in my personal benchmark. Of course, it's subjective, but the difference was bigger than I expected.

Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

The single greatest "test or check" of being truly free is to ask yourself... if you're allowed or you allow yourself to ask questions from the authorities. Be it your religion, government, management, your beliefs, any idea you might've become slave to unknowingly.

Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

Bruh! I don't know a single person who's using loveable, bolt, windsurf, threads, and yet they keep claiming trillion of users.

Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

The more we try to understand supermassive blackholes, the more physics it breaks. One the benchmarks of AGI should be the to figure out blackholes.

The more we try to understand supermassive blackholes, the more physics it breaks. One the benchmarks of AGI should be the to figure out blackholes.
Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

Every time I open X, I see something relatable, something of my interest, and the account is followed by the same set of people. There's definitely an Indian tpot group with same interest and intellect level.

Golden_Iris⚡️ (@golden_iris7) 's Twitter Profile Photo

Everything in AI is a hack that worked, not elegant theory. ReLU, dropout, batch norm, these aren't derived from first principles but trial & error successes. Even transformers' QKV mechanism was a "let's try this differentiable trick" guess that exploded in power.