Archie Sengupta
@archiexzzz • 69,542 subscribers
researcher @midcenturyai (robotics & world models) // researcher @sarvamai // pre & post-training // prev: @atomicworkhq, 5x yc dev
Shorts
Videos
Are you kidding me? This is the best website design experience I've had in a long, long time. Shopify just won my heart with this new Winter Edition web design! Seriously, the bar for good design is so high right now. With all the AI slop out there, the only word-of-mouth you can rely on is having great design taste and building a cult around it. Just take Prime Intellect - ultimate design taste. Why wouldn't I send this to my friends or post it online organically? The moat is good design, folks - it's so important with all the AI slop we're seeing today. tobi lutke please give your design team a big big raise!
Archie Sengupta250,690 views • 8 months ago
8-year-old founder at south park commons. this is who you are competing with btw
Archie Sengupta104,543 views • 10 months ago
If I were an LP in this fund, I'd be deeply worried seeing this. No idea what was the reason of posting this on socials instead of keeping their funding thesis with themselves(ragebait?) There's a reason why some of the best founders I know actively seek out American VCs and are cautious about certain 'lala' Indian VCs. And to be clear - I'm not generalising. Several Indian VCs I've personally met are exceptional, with remarkable track records and portfolio companies built by founders < 30. A 24 or 25-year-old founder: their only obsession is solving a problem and building. They're not thinking about ordering diapers for their 2 y/o on Blinkit. No dependents, no financial obligations pulling their focus - just a singular, relentless drive to crack that one problem their company exists to solve. Does experience matter? Absolutely. But does that mean we should actively discourage younger founders from even trying? that's just bad taste from a venture capital firm. Hear what Jeff Bezos has to say:
Archie Sengupta41,519 views • 3 months ago
Huge congrats to Skyroot Aerospace & team for successfully taking Vikram-1 Test Flight-1 to orbit. India's first privately developed orbital rocket has completed its final burn and injected its payloads into a ~450 km orbit, making India the third country in the world with private orbital launch capability. 🇮🇳🇮🇳
Archie Sengupta24,146 views • 2 months ago
Every now and then, I come back to this interview of Shah Rukh Khan on how to be successful(for the younger generation): "you can't be restful and at peace if you want to be successful and it's the truth, it's god's honest truth." "jo bhi apko bole aram karo, zindi jo h is beyond work and all, its good that you donot want to be successful. i feel envy and envious of people who relax and rests. i feel jealous of people who dont want to be successful". learn from the greatest. always.
Archie Sengupta65,956 views • 1 year ago
vLLM fast inference running on monte carlo synthetic data generation with: > peak generation throughput of ~ 23k token/s & avg of 20k token/s > ~200 reqs/s > Qwen/Qwen2.5-0.5B-Instruct > on 1x Grace Hopper 200 - 480 GB(~96GB HBM3) vllm config: --max-num-seqs 512 --chunked-prefill-enabled (for better throughput) --dtype float16: (half precision for mem efficiency) --gpu-memory-utilization 0.95(for better KV Caching)
Archie Sengupta26,506 views • 8 months ago
Cursor posted an open problem - Character Prefix Conditioning on their blog post. This is my attempt to solving the problem using DFA + trie + SmolLM-135M. LLMs generate tokens, not characters. But we type characters. If the user has typed a few characters in the current word(e.g. pri) we want the model's completion to begin with those exact characters. To solve this we: > build a trie of vocab tokens > track prefix consumption with DFA state(index i) > at each step, allowed tokens = {tokens that are prefix of R} U {tokens that start with R} > force deterministic transitions if only one option, else sample with masked logits. sample video on how it works. ofc it's not perfect, it's just a 135M model, you can choose whatever model you want :))
Archie Sengupta24,973 views • 1 year ago
No more content to load