head of growth + devrel blokcapital | building gtm engines with ai
Joined July 2024
- Tweets18.1K
- Following1.1K
- Followers1.8K
- Likes14.6K
Pinned Tweet
Finding a co-founder shouldn’t take months
So I fixed it
Meet cofoundr
> Find someone to build with in 60 seconds
> No cold DMs
> No endless scrolling
> Just real builders, ready to start
It’s live. It’s free
Try it before your next idea stays just an idea
Link in comments
just watched #Ramayana
nd am so proud for this
this is not just vfx, its a visual poetry
did u expect this kind of graphics from an indian film?
revolut moved 2,000 engineers off claude code and codex
and cut its ai coding bill by ~90%
they didn't find a cheaper plan. they downloaded a model and ran it themselves
> GLM, an open-weights model from a chinese lab
> runs on their own servers in the US, nothing goes back to the lab
> 56 nvidia B200s serve all 2,000 engineers
> works out to ~$0.30 per million tokens
the part most people miss: they didn't fully drop claude
it's still there, but only for the hardest tasks. the cheap model does everything else
i do the same in cofoundr. every call tries a 4¢-per-million model first and only steps up if it fails
how it actually works, ep 8 👇
detail i cut from the video: GLM is text-only
so the router also sends anything with images (screenshots, diagrams) to outside models like gemini and gpt
self-hosting only covers the work your model can actually do
source: revolut's infra lead talking to the register. the 90% is the inference bill. full hardware + ops cost isn't public, so read it as "per-token cost dropped 10x", not "total ai spend dropped 10x"
52
Rohit Purkait retweeted
i currently write scripts for my videos
the problem is
- sometimes it doesn’t feel natural
- takes quiet some time to memorise the script
do you guys think it’s better to sit down and just talk about whats in my mind?
anthropic just taught claude to dream
and on one memory test, it nearly doubled a model's score without changing the model at all
so in this video, i've explained exactly:
> why AI agents forget everything between chats
> how an older claude wrote 413 memory files and got 1 pokémon badge, and a newer one kept 10 and got 4
> what "dreaming" actually does between sessions
> how one wrong note sent claude round the same route 3 times
> the 200-line limit in claude code's MEMORY.md, and how to use it today
follow along for more AI, explained from the source.
i currently write scripts for my videos
the problem is
- sometimes it doesn’t feel natural
- takes quiet some time to memorise the script
do you guys think it’s better to sit down and just talk about whats in my mind?
anthropic just taught claude to dream
and on one memory test, it nearly doubled a model's score without changing the model at all
so in this video, i've explained exactly:
> why AI agents forget everything between chats
> how an older claude wrote 413 memory files and got 1 pokémon badge, and a newer one kept 10 and got 4
> what "dreaming" actually does between sessions
> how one wrong note sent claude round the same route 3 times
> the 200-line limit in claude code's MEMORY.md, and how to use it today
follow along for more AI, explained from the source.
Rohit Purkait retweeted
anthropic just taught claude to dream
and on one memory test, it nearly doubled a model's score without changing the model at all
so in this video, i've explained exactly:
> why AI agents forget everything between chats
> how an older claude wrote 413 memory files and got 1 pokémon badge, and a newer one kept 10 and got 4
> what "dreaming" actually does between sessions
> how one wrong note sent claude round the same route 3 times
> the 200-line limit in claude code's MEMORY.md, and how to use it today
follow along for more AI, explained from the source.
anthropic just taught claude to dream
and on one memory test, it nearly doubled a model's score without changing the model at all
so in this video, i've explained exactly:
> why AI agents forget everything between chats
> how an older claude wrote 413 memory files and got 1 pokémon badge, and a newer one kept 10 and got 4
> what "dreaming" actually does between sessions
> how one wrong note sent claude round the same route 3 times
> the 200-line limit in claude code's MEMORY.md, and how to use it today
follow along for more AI, explained from the source.
trade elysium on day 1 with 0 fees
@BitgetWallet is backing @Enter_Elysium from launch
ZERO trading fees for the first 5 days
elysium is the newest corner of the hyperliquid ecosystem, and it's early
get in before it gets crowded.
trade more, pay zero
open bitget wallet and be an early elysium trader 🔥
🤝 Paid partnership
Rohit Purkait retweeted
happy late diwali guys
was so busy past week, totally exhausted
now gonna take some rest n chill
Rohit Purkait retweeted
my 2cents after winning 20+ hackathons this year
so in this video, i've explained exactly:
> why the best ideas don't come from brainstorming
> chithi, cofoundr and tastemaker, and the problem behind each
> the two apps i built that nobody picked up, and why
> the filter i check before i build anything
> the extra filter that helped me win 20+ hackathons
got a question you want me to answer?
drop it in the comments, follow along, and share this with a friend who keeps saying "i have an idea"
Rohit Purkait retweeted
first in my bloodline to cross 1.8k followers
grateful to everyone who have been part of the journey
thanks for supporting my experiments and showing me that the decision i took at the beginning of the year is worth it
time to go crazier
my 2cents after winning 20+ hackathons this year
so in this video, i've explained exactly:
> why the best ideas don't come from brainstorming
> chithi, cofoundr and tastemaker, and the problem behind each
> the two apps i built that nobody picked up, and why
> the filter i check before i build anything
> the extra filter that helped me win 20+ hackathons
got a question you want me to answer?
drop it in the comments, follow along, and share this with a friend who keeps saying "i have an idea"