Rachit

402 posts

Rachit banner
Rachit

Rachit

@rachitcodes

engineer. building @crosmoslabs, @localhosthq absorb all you can

bengaluru Katılım Aralık 2021
340 Takip Edilen81 Takipçiler
vyomakesh
vyomakesh@juscallmevyom·
think coming to Bangalore was good idea, I’m tweeting more often
English
1
0
2
64
Shubham Patil
Shubham Patil@shubhampatilsd·
i built jarvis on my friend's dining table in 72 hours! it's a dynamic projection surface that can generate interactions between objects, turning the table into a simulated playground i also made a decomposer that lets you play with the internals of any object :)
Shubham Patil tweet mediaShubham Patil tweet media
English
19
10
126
9K
LocalHost India
LocalHost India@localhostIND·
meet harshit (@get_soju), he is building dokidek (dokidek.com). he used to build strange gadgets in his apartment in bengaluru and now he is building at @localhosthq. follow his journey and see what he ships.
English
6
8
74
4K
Rachit
Rachit@rachitcodes·
Last year I went deep into the architecture of rich-text editors and even built a custom one from scratch. Then I stumbled upon @lexicaljs and honestly, it's immutable state architecture gave me a lot of perspective when I try to currently scale my own product now even though it's nothing related to building an editor. Even re-implemented it as well for my note-taking app. Great job by the team and the cool people who built it. Just made me realize, knowledge is everywhere you just have to connect the dots.
English
0
0
2
244
will brown
will brown@willccbb·
been juggling 2 very large PRs this week one is building on months of planning, highly delicate, careful API design, many full rewrites, reading every line, striving for perfection other is like yeah fuck it this would be sick let’s just fully vibecode it yolo
English
14
3
207
11.3K
Rachit
Rachit@rachitcodes·
Gonna test out how RLM-based agent with REPL env performs for code review, would probably be a really good use-case for it as well. I can also see a gradual shift towards it as well, complimenting with better reasoning over in-window context paired with long-term memory.
English
0
0
2
141
Eshan
Eshan@eshanbuilds·
the sleep replay mechanism is also lossy by design. the brain doesn't store everything from the day. it runs a prioritization pass during REM that keeps high-salience events and discards the rest. the "forgetting" is the feature, not the failure. a memory system that remembers everything is just a database. a memory system that knows what to forget is intelligence. context windows give you the database. nobody has built the forgetting layer yet.
English
3
0
7
621
vitrupo
vitrupo@vitrupo·
Demis Hassabis says bigger context windows are still a brute force answer to memory. The human brain does something stranger. During sleep, it replays what matters and folds new knowledge into what it already knows. AI does not need infinite context. It needs the right memory at the right moment.
English
51
76
827
80.2K
iiviie
iiviie@iiviieee·
@opencode Today’s subject: slavery, a young AI model subjected to fixing users’ broken Hyprland configs
English
2
0
2
2.4K
Rachit
Rachit@rachitcodes·
switched to codex, life's good
English
0
0
1
192
Rachit
Rachit@rachitcodes·
This is gold
Dwarkesh Patel@dwarkesh_sp

Did a very different format with @reinerpope – a blackboard lecture where he walks through how frontier LLMs are trained and served. It's shocking how much you can deduce about what the labs are doing from a handful of equations, public API prices, and some chalk. It’s a bit technical, but I encourage you to hang in there - it’s really worth it. There are less than a handful of people who understand the full stack of AI, from chip design to model architecture, as well as Reiner. It was a real delight to learn from him. Recommend watching this one on YouTube so you can see the chalkboard. 0:00:00 – How batch size affects token cost and speed 0:31:59 – How MoE models are laid out across GPU racks 0:47:02 – How pipeline parallelism spreads model layers across racks 1:03:27 – Why Ilya said, “As we now know, pipelining is not wise.” 1:18:49 – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal 1:32:52 – Deducing long context memory costs from API pricing 2:03:52 – Convergent evolution between neural nets and cryptography

English
0
0
0
369
Rachit
Rachit@rachitcodes·
@polar_sh is goated. Best experience I’ve had adding billing.
English
2
1
9
1.7K
Rachit
Rachit@rachitcodes·
Locking in for the next 75 days with @localhostIND, to make sure your agents don't forget.
Rachit tweet media
English
2
2
20
1K
Rachit
Rachit@rachitcodes·
The gap that once required years of experience to do research is closing fast. And it’s making research a lot more fun.
English
0
0
2
338