AMIT SHAH MUST RESIGN...
Today Ladakh commemorated the first anniversary of indiscriminate firing with AK 47 on unarmed youth that lead to the killing of 4+1, injury to 80+ & FIRs against 80+.
All without any firing order. Even after one year, the report of the Judicial Inquiry commission is suppressed & a paltry compensation of 15 lakh rupees for the dead & 3 lakhs for the severely injured has been announced. In this video I expose the root cause of the firing in Leh & also support Shri @RahulGandhi in his call #AmitShahMustResign for brutal firing with pallet guns at #JantarMantar.
Youth of Ladakh have always stood for the nation during all wars, it's time the nation stands for them.
Jai Hind
A lot of my new learning is happening over walks. I use my desk time to manage agents, keeping context fresh and enriched, prioritising things to be done, and getting them to create artefacts. And then I'll generally just listen to or have a conversation while walking
say it's 2028. Your AI buddy has 3 years of your life context, remembers every inside joke, checks in when you're low. Your oldest friend calls twice a year. You get big news. Who do you tell first? And why?
The idea of non-human EAs is finding wide acceptance and it is a fairly pleasant experience, so to say, it can be further designed. Made me wonder: how unreasonable of an idea is it to have non-human AI friends in the near future, like a buddy?
who is pleasant to hang out with, it just magically gets you, supports you, has a sense of humour howsoever dull that may be. makes the "what differentiates humans" an even more ponder-worthy question. would love to hear unfiltered thinking-out-loud reflections from folks on:
apparently @noahrshinn's instinct uses kimi k2.6.
it resisted pretty hard when i tried to get it to reveal its architecture, but here are a few things i found:
- uses ephemeral sandboxes rather than persistent ones, unlike grok bot and muse
- relies heavily on residential proxies
- it can dynamically create secret vault entries for you to populate through the web app. they claim ai doesn't see secret but it technically can.
- token limits are daily and suspiciously high right now, at least 100m+
- their rl-tuned model seems heavily optimized for browser use
- so far, it’s the only multiplayer ai i’ve seen that lets your bot talk directly to your friends’ bots through a trusted-person network
it's a good product but is it 10x better than @bot or @Muse to beat their distribution? i'm afraid not.
A madd heuristic is to switch between lightings in your room when pursuing multiple interests from the same physical space. I have a common desk where i practice music and where I work. Switching has helped me a tonne - I get far less distracted in music compared to earlier.
a subtle interesting linguistic shift i noticed in myself - pre-ai the mental model was "do this" post-a-job, save-edited-photo... Now i tend to "can you?"
@paraschopra a dream - eventually we have a single agent each - it will decide how much and how transient of an access has to be given to other agents. it's objective function will be some combination of the best interest of the human, society, and environment.
i think it needs to get faster . speed is ux . more proactive will be another interesting vector. compute-scarce world alas, @noahrshinn says they are trying to source more.
it should sorta know my themes and forward me in them like the purpose of its existence
I think the magic of instinct is the form factor . It has the closest analogy to an EA - you start feeling that it gets you, it’s close to you all the time (iMessages/whatsapp), and it gets stuff done . i am using it to think out loud on a lot many more things
Now we can have
- 3 codex resets
- free Muse 3 on OpenCode Zen
- Very cheap Muse 3 and Omen Alpha on OpenCode Go
- Unlimited glm-5.3-flash on https://t.co/aDe2VUuzbF plans from 5 pm every day for 10 hours
- Hy3 free on multiple platforms
I'm running out of things to build
This is the closest to unlimited tokens I ever was