mayank (@SimplerMayank) profile picture

just another engineer

hyd/del
Joined July 2022
Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)Photo by mayank (@SimplerMayank)
I reverse engineered Pocket and made it 10X better. Right now: the app is full of ads, the microphone is low quality, it constantly misspells peoples names, and has no agentic features. I changed that.
134
94
54
1,990
170,485
hey @rumik_ai i've been following you guys for months, and i want to work with you guys and after overthinking everything...here's my pitch
25
7
9
291
21,085
shoutout to my friend Vijay, who helped me shoot the video here's his instagram, give him a follow- instagram.com/allabout.vijay
4
1,897
introducing common! a whatsapp group chat for your friends and agents. i made it for myself to share a clean space for people i care about and agents we use. - invite friends by email - let agents connect to space with MCP - free, private, open source. try at: usecommon.​app
1
4
491
introducing Mania.inc ✨ the greatest AI company ever built to waste your time, instead of saving it. may the fun begin.
110
58
46
804
176,653
did a v1 of game theory with jev in v0, once jev defected, it never went back to cooperating. so in v1, i added two more scenarios to make it cooperate or defect in the 1st round it rarely went back to cooperating (12/720) and still beats other strategies in all 3 scenarios
i tested game theory with jev. put it into prisoner's dilemma, and it beats tit-for-tat as well. across 40 matches, it averaged 2.28 points per round and cooperated only ~20% of the time. its strategy resembled Always Defect and Grim Trigger
1
1
313
jev had the highest average in all 3- - picks its own opening: 2.83 pts/rd vs 1.56 for the next best (tit-for-tat) - cooperate first: 3.03 vs 3.00 (always cooperate) - defect first: 2.79 vs 1.00 (always defect), and it never lost a match
174
video editing using jev in real time
built a deterministic editing engine for snip and it now parses and validates JSON output to apply edits this unlocks many use cases, one of which is to let your agent edit videos for you through the local MCP "you can now use your agent to edit videos in snip"
2
7
217
built a deterministic editing engine for snip and it now parses and validates JSON output to apply edits this unlocks many use cases, one of which is to let your agent edit videos for you through the local MCP "you can now use your agent to edit videos in snip"
i was looking for a simple way to edit screen recordings but existing ones were either bloated or included watermarks. so i built snip - a simple editor for small potato stuff - split into multiple clips, trim them, adjust speed or zoom in - export video, free and no watermarks
1
1
332
i was looking for a simple way to edit screen recordings but existing ones were either bloated or included watermarks. so i built snip - a simple editor for small potato stuff - split into multiple clips, trim them, adjust speed or zoom in - export video, free and no watermarks
1
1
3
346
i tested game theory with jev. put it into prisoner's dilemma, and it beats tit-for-tat as well. across 40 matches, it averaged 2.28 points per round and cooperated only ~20% of the time. its strategy resembled Always Defect and Grim Trigger
1
1
1
4
487
- if Jev defected first, it kept defecting - if both players cooperated first, Jev kept cooperating - if Jev cooperated first but the opponent defected, it switched to defection in round 2 and never switched back
1
2
81
mayank retweeted
New experiment: json-render + jev The future Generative UI is instant Your components, your actions, your design system Rendered in milliseconds
223
507
265
7,787
1,348,844
in package version caret (^) usually pins the major version and tilde (~) pins the minor version caret is less restrictive than tilde, but it gets tricky with a zero version component ^0.5.0 - pins 0.5 and allows only patch updates ^0.0.5 - effectively pins 0.0.5
1
2
100
surprising thing was ~0.0.5 allows newer 0.0.x versions (<0.1.0) and for 0.0.x, ~ is actually less restrictive than ^
1
82