Optimize your life to impress peers -> get stuck in a local maxima defined by others.
Optimize your life around doing what you enjoy the most -> keep climbing towards global maxima solely defined by you.
Google has found something worst than AI hallucinations.
they call it “Metacognitive Failure.”
when an LLM hallucinates a fact, it's a data error. bad memory retrieval, wrong weights, fine. you can fact-check it.
metacognitive failure is a structural psychological defect in the model's architecture.
according to the research, current frontier models exhibit massive gaps in their internal self-monitoring:
the supreme confidence trap: they routinely hallucinate with maximum, bulletproof confidence. they sound just as sure of themselves when they are completely wrong as they do when they are 100% right.
the boundary blindness: they have zero internal mechanism to recognize their own knowledge boundaries. they blindly step off cliffs because they can't sense the edge.
the calibration mismatch: there is a complete disconnect between what the model actually "knows" internally and what it expresses in its output.
the fix: reinforcement learning with metacognitive feedback (rlmf)
the researchers operationalized a new technique to force models to face reality. instead of just rewarding the model for getting the right answer, rlmf grades the model on how accurately it evaluates its own performance and uncertainty.
by aligning a model's expressed confidence with its actual intrinsic uncertainty, they managed to massively improve model calibration without dropping raw accuracy.
as we shift into an agentic era where ai systems are running code, managing infrastructure, and making automated decisions without humans in the loop, a model that doesn't know what it doesn't know is dangerous.
An IIT Bombay professor wants to build cities in space!
This week, investors gave his startup ₹100 crore to begin with something slightly less impossible: repairing satellites without bringing them back to Earth.
Professor Arindrajit Chowdhury’s idea began far below orbit, inside a Mumbai traffic jam. It had taken him nearly an hour to travel four kilometres.
After more than a decade teaching mechanical engineering at IIT Bombay, Chowdhury found himself wondering what would happen as cities became even more crowded.
Where would the space, energy and resources come from?
His answer sounded absurd: move some of civilisation’s infrastructure beyond Earth.
In 2022, Chowdhury and his former student, propulsion expert Dr. Tausif Shaikh, founded InspeCity. Inspired partly by physicist Gerard O’Neill’s vision of rotating space habitats, they wanted to build factories and, eventually, human settlements in orbit.
Chowdhury took a year’s leave from IIT Bombay to pursue it.
But they did not begin by building a city. They began with a propulsion system roughly the size of a small tiffin box.
The immediate problem was already circling Earth. A satellite costing millions can become useless when it runs out of fuel, loses manoeuvrability or develops a fault. Sending astronauts to repair it is extraordinarily difficult and expensive. Abandoning it adds another dead object to an increasingly congested orbit.
InspeCity wants to become the mechanic that travels to the satellite.
Its technology stack includes GITA for propulsion, CHAKSU for sensing and navigation, RAMA for robotic manipulation, and SPARSH for docking and transferring propellant in orbit.
Together, these systems are designed to let spacecraft inspect, approach, refuel, repair or safely deorbit other satellites.
This week, InspeCity raised ₹100 crore, approximately $10.5 million, in a pre-Series A round led by Speciale Invest and investor Ashish Kacholia.
The company now plans four orbital missions over the next 12 to 18 months.
The first two will test whether its spacecraft can approach another object safely. A later mission will attempt what Chowdhury calls a “handshake” in space. The mission after that is expected to attempt docking.
InspeCity has not repaired a satellite in orbit yet.
That is precisely what makes the next chapter important.
A professor who once spent an hour travelling four kilometres is now attempting a far harder journey: closing the final few metres between two spacecraft moving through orbit.
the Lionel Messi of Google just left
> be Jeff Dean
> join Google in 1999. Employee #30
> build mapreduce, then bigtable, then spanner
> don’t stop, co-found google brain
> build tensorflow
> still not enough, build the TPU
> any one of these is somebody's entire career
> engineers start telling jokes about you like you are a folk hero
> one of the most cited names in AI and distributed systems
> stay 27 years at Google, have a massive impact on computing
> get the startup itch again
> walk out with sanjay ghemawat, oriol vinyals and quoc le
> start a new company, get funded day 1, obviously
> google invests in it and becomes your cloud partner
> alphabet loses close to 5 percent the same day
> ~200B$ market cap lost because of 4 engineers
wow.
silicon valley is the greatest show on earth!
We are real time seeing software saturating the total global available human attention
It's incredibly hard to launch pure software consumer products today because people are just tired of "yet another app"
There's trillions of lines of code yet to be written in the world, despite that though. Let no one convince you otherwise
Saw a broken speed breaker on road and got a crazyy thought
One of the ways to get India to follow lane discipline could be strategically placing cuts in the speed breakers such that it doesn't bump those who follow lane discipline - and watch everyone get in line 😄😄
All of us have seen the new age startups around us grow, we totally miss seeing how the foundation of today's India was laid by all such upstarts like Titan, brilliant leaders like @XerxesDesai all of their contributions upon the shoulders of which today's India grows!! ♥️
@prakdadlani But how does it help? If a manufacturer like you believes in a product you'll buy/invest in the mould yourself right - curious to understand why you want a loan for this? What interest rate are you willing to pay? And you'd be okay tying your credit rating to it right?