Long horizon agent building AI native wealth management @ Atyx | I build things | ex-Product @ PhonePe

Joined February 2009
What makes models seem "good" at thinking? They can be great at picking a set of arguments and stitching them cohesively, glossing over all dissonance
119
First they came for the software engineers, I said nothing for I was not a software engineer. Then they came for the mathematicians, I said nothing for I was not a mathematician. Then they came for the product managers and I said "make sure you get all the product managers"
39
270
8
5,773
117,226
"Go through my pending tasks, present them to me card by card in an HTML file, make it into a game with choices like "Pick up / Delegate / Defer / Question why it needs to be done / This doesn't need to be done "
Me, when I look at all my pending tasks:
1
8
527
What are a few great non-fiction non-management books you recommend? Maybe something that reads like anecdotal history of things / places / companies etc (or anything else that blew your mind)
8
3
731
These sections in the book store keep growing bigger 馃榿
3
14
716
This is @cursor_ai preventing me from thinking through bug fixes on my own code base because the prompt is "sensitive". Things seem to be getting worse across the board
1
209
Ankit Daftery retweeted
We have two rules for code going to production @PhonePe : 1. You own the code you commit 2. You partially own the code you review and approve How the code is generated is irrelevant. Been working out fine so far for the most part.
77
62
13
2,270
92,395
Too early to tell but I get a strong feeling that small language models are going to eat LLMs for lunch. Deep inside, everyone's kinda tired of these foundational models and the companies that make them.
1
246
Ankit Daftery retweeted
Jev is going to result in an explosion of small, specialised models. And I'm absolutely here for it.
11
11
5
177
4,755
Ankit Daftery retweeted
The Jev thing that is blowing my mind right now is how we can invert the harnesses. Classifiers on the outside loop. Generative on inside tools. Token cost and latency are both going to go way down.
18
34
4
467
20,473
DGX Spark is a gateway drug to being a data center tycoon, kids
Another 2 or 3 DGX sparks will fix me. For a while at least, I guess.
11
542
Spent the evening fixing MAE because a power cycle after 3 months of uptime corrupted it. Muse Spark + Opencode debugged the OS image, made a surgical recovery plan, executed it, and created a backup with very minimal nudging. I am back to not having to read all my email 馃檪
14
421
This is a paid tweet, but in a different kind of way. I had to pay to try this out 馃榿 Very very cool implementation of agentic payments, something I would truly feel comfortable using and giving access to my payment credentials
Hey chat, I鈥檓 hungry.
4
2
1
12
3,650
@npulyani get a demo of this from @Sengupta
1
213
Ankit Daftery retweeted
Hey chat, I鈥檓 hungry.
11
12
6
36
13,647
Ankit Daftery retweeted
IndianRailwayBench - where we rank LLMs on their ability to book tatkal tickets from ircrc
54
83
17
1,877
46,247
Hypothesis : If you're finding a document hard to read and follow, it's almost certainly generated with Claude. Corollary 1. The sender has probably not read it themselves 2. The sender won't stand by what's written in it 3. You might as well discard it
1
1
1
17
826
Ankit Daftery retweeted
That鈥檚 the load-bearing insight. The quiet part is: this isn鈥檛 a bug, it鈥檚 a feature. Zoom out. The takeaway? It just works. Seamlessly. Effortlessly. No notes. Rest is noise. And honestly? That鈥檚 the whole ballgame. Let that sink in. It鈥檚 not X鈥攊t鈥檚 Y.
157
47
29
1,172
89,036
Ankit Daftery retweeted
i do not know who needs to see this (but i have a feeling most of you do)
5
15
1
234
11,069