ManIT

building vektori (continually improve models)
cs student • cofounder • the residency delta
ai memory systems, distributed systems, rl

[ManIT @AlexQHunterz]

@indilingo @_anshulgarg @aayushmor_13 @_im_nitesh @jaygala223 @GauravSarkar99 Yoooo crazy

[ManIT @AlexQHunterz]

@punit_arani Fair enough another thing is that do u fair enuf btw do you think it's possible for Opus to create a task that Opus itself fails at entirely synthetically?
(If that happens its rsi lol)

[ManIT @AlexQHunterz]

@danielrupawalla Fair enough makes sense.

[ManIT @AlexQHunterz]

@danielrupawalla but what about certain high volume tasks for internal models?
Tasks that dont require lot of intelligence(which is why they use internal models) or require the data not be exposed.
What about those?

[ManIT @AlexQHunterz]

there is no greater danger than playing it safe.

[ManIT @AlexQHunterz]

@SarvamAI @lightspeedvp @GrowthX_Club

[ManIT @AlexQHunterz]

At the sarvam hack here
Just submitted under the

[ManIT @AlexQHunterz]

@gagan_builds Submission deadline lmao

[ManIT @AlexQHunterz]

@ilavanyajain all in bro

[ManIT @AlexQHunterz]

Relentless pursuit all the way, go all in or don't do anything at all

[jo johnson @josbjohnson]

one common trait I find in high achieving individuals is urgency. it looks almost like impatience until you understand it is closer to reverence, a deep refusal to waste the one thing they can’t buy more of. the alarm goes and they rise. the message is answered while it is still warm. the small task is done before it can find a shelf to hide on. they commit and move in a single motion. an idea sparks and they chase it that day, before the world talks them out of it. they run at their own lives like something is at stake, because they have decided something is.

[ManIT @AlexQHunterz]

Thanking @googlecloud @grayscale_vc @testsigmainc for giving us this oppurtunity.

RSVP here
https://t.co/925PTWX18k

[ManIT @AlexQHunterz]

Everyone's solving agent memory by stuffing more context into a bigger model.
We trained a small model instead scored by an RL environment to decide what it keeps, drops and resurfaces for one workflow.
Proto live at our stall, Agentic Summit BLR × @Magicball, July 22.

[ManIT @AlexQHunterz]

@JunaidAckroyd https://t.co/ngC3wyHmDw

[ManIT @AlexQHunterz]

@llmluthor 100x engineer

[ManIT @AlexQHunterz]

@MaximeRivest This is so cool

[ManIT @AlexQHunterz]

@johnowolabi_ Hey would be interested to work as a dev.. would love to have a chat.

[ManIT @AlexQHunterz]

@alokbishoyi97 Interested

[ManIT @AlexQHunterz]

@llmluthor Hmm what's the funniest/weirdest reward hacking scenario untill now