sumit

2.9K posts

sumit banner
sumit

sumit

@sumitdotml

swe. learning how to think in matrices

日本 東京 شامل ہوئے Eylül 2022
524 فالونگ9.9K فالوورز
sumit
sumit@sumitdotml·
@weeyev lmao you were waiting for my quote???? 😄
English
1
0
1
26
sumit
sumit@sumitdotml·
@astrodotbuild you guys need to come to Tokyo I’d love to share my sites with you
English
1
0
1
40
Astro
Astro@astrodotbuild·
Who’s in London next month? Come hang out with the Astro core team! Sign up now: luma.com/lzh944vx
Astro tweet media
English
7
11
79
7K
sumit
sumit@sumitdotml·
@dejavucoder been pretty unreliable for me as well, I resorted to tailscale + termius & now it’s been super smooth
English
0
0
2
222
sankalp
sankalp@dejavucoder·
bruh claude code remote control session does not connect half the time. i thought my wifi had issues.
English
10
1
52
10.2K
sumit
sumit@sumitdotml·
fifth time just today that claude code has failed to use the first bullet point of a skill despite explicitly being told to use that skill I'm thisss close to cancelling my subscription
English
1
0
3
855
sumit
sumit@sumitdotml·
who am I to even tell people what to do & what’s right or wrong but Mitchell Hashimoto announcing that he’s joining that company has disappointed me a lot
English
10
3
284
29.8K
sumit
sumit@sumitdotml·
one more g*tack psychosis tweet on my timeline to finally send me over the edge
English
1
0
7
520
sumit
sumit@sumitdotml·
@Asura1612913 this is hilarious I remember seeing his account saying he was 16 when he was at tilde, I didn't know he's still this age lol
English
0
0
0
68
Asura
Asura@Asura1612913·
@sumitdotml *16, anyways the paper was really enjoyable to read
English
1
0
0
83
sumit
sumit@sumitdotml·
@thsottiaux hello good sir any plans on this tier at oai?
English
0
0
0
205
sumit
sumit@sumitdotml·
this happens on a daily basis now, the $100 codex plan can't come soon enough
sumit tweet media
English
1
0
4
647
sumit
sumit@sumitdotml·
@yeginbay yup I can do it now thanks
English
0
0
1
35
Alisher
Alisher@yeginbay·
@sumitdotml looks like this was fixed just now! u can try again, i was able to login using URL that /login provides
Alisher tweet media
English
1
0
1
76
sumit
sumit@sumitdotml·
this is getting ridiculous now I've had to re-login to claude code multiple times these past few days due to their repeated api error the agents that are launched during /loop have stayed idle multiple times and blocked my cron jobs when I haven't checked, and now this oauth error that does not even allow me to login. this has killed my current /loop cron as well :( so irritating
sumit tweet media
English
4
0
15
1.5K
himanshu
himanshu@himanshustwts·
Career update: Excited to share that I have joined the incredible team at @smallest_AI to work on Research x Devrel! The team is cooking incredible small + efficient multi-modal models and it feels like an exciting time to push the frontier on scale!
himanshu tweet media
English
205
32
1.8K
59.4K
sumit
sumit@sumitdotml·
sumit tweet media
ZXX
4
5
115
5.5K
Kimi.ai
Kimi.ai@Kimi_Moonshot·
Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation. Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers. 🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth. 🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale. 🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead. 🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains. 🔗Full report: github.com/MoonshotAI/Att…
Kimi.ai tweet media
English
326
2K
13.4K
4.8M