Dear frontier labs:
Please build me an agent that actually passes the brown M&M test.
I’ve been deep in OpenClaw + Claude Cowork/Code for months, with obsessive memory management and daily byterover curation.
Yet important details, custom rules, and hard-won context still quietly vanish or get ignored.
Benchmarks look impressive. Real-world reliability? Brown M&Ms everywhere.
We don’t need a smarter model. We need one that actually reads the damn bowl.
Who’s solving this?
After April 28, Science is Cool 16 is heading to Hawaiʻi on July 23!
Active lava researchers, Mauna Kea astronomers, and marine biologists from the Pacific. Free for science teachers. More soon!
https://t.co/aVAV4aEAF0
Sal Khan figured out why AI tutoring flopped with students: they don't know how to ask good questions.
Here's the deeper layer: AI rewards bad prompting with seemingly good responses, creating a trap.
An AI model is a mirror of your own clarity of thought. Vague in, vague out. Learning to write clearly is even MORE important now, not less.
https://t.co/KOYVyprh6A
@khanacademy@salkhanacademy
@chamath You’re overthinking it. Glasswing is a beta test. Eng wants to get buggy sw in front of their best customers who’ll trade bugs for early access. Dario’s “terminator event horizon” is merely mkt air cover
Congrats on the Artemis II launch last night! Four astronauts on their way to the Moon! 🚀🌛
This fall, with help from @voyagertech_, a PocketLab sensor goes to the ISS National Lab! @ISS_CASIS
Students on Earth will run the same experiments as the astronauts. Same sensor, same data, different gravity.
We're not going to the Moon (yet), but we're getting pretty close! 😄 @NASAScience_
Steve sent photos of his studio! So cool to see what @SteveSpangler's set up looks like from behind the camera 🎥
This is where we're going to stream "The Science of Magic" live on April 28th.
Sign up now - it's FREE!! https://t.co/gPmLSqgWAe