Tesla owners are starting to realize that despite paying $8K for FSD back in 2017 they are never going to get it as the latest software no longer runs on their hardware. www.wsj.com/business/aut...
Today, inundated by reporters wanting to know implications of humanoid robot winning a half marathon. NONE! As a kid read about a Chinese American serviceman beating a computer with an abacus. 1000's of humans separated by a fence from robot running in the same sized lane. Wheels, people, wheels.
Blue Origin successfully landed booster "Never tell me the odds" for the second time. Big success. But something awry on second stage and payload deployed in a off-nominal orbit. No word yet on how badl and whether on board thrusters have enough fuel to accommodate. spacenews.com/third-new-gl...
Blue Origin launching used NewGlenn Sunday--paying customer. Blue Moon Mk1 now getting stacked; 3 tons of cargo to Moon S pole in 2026. 2027 plan: human rated Mk2 demo dock with crewed Artemis in Moon Orbit; aim to land crew in 2028. No in space refueling. www.scientificamerican.com/article/nasa...
I think they found a shoe that fits.
The need to believe in the glorious AI future is strong indeed. Remember earlier this week when the NYT touted the $1.8B AI startup with just two employees? It is all a completely faked front end with 800 fake doctors connected to an existing platform. futurism.com/artificial-i...
Learning is a suitcase word (see Marvin Minsky). Learning about stop signs is a suitcase problem--very different for fixed stop signs, hand held stop signs, and stop signs on school buses. Waymo is not learning the latter in Austin at all well. Critical safety issue. www.wired.com/story/a-scho...
Gill Pratt, Chief Scientist at Toyota, ran the DARPA Robot Challenge in 2015. First time walking humanoids were on the world stage. Earlier he ran the "Leg Lab" within the then MIT AI Lab. Read where he thinks we are now in our robot future (spoiler: its nuanced). spectrum.ieee.org/humanoid-rob...
Artemis II launch window opens at 6:24pm ET tomorrow (April 1st), sending four astronauts around the Moon--just over 53 years since last human visit. Technology and capabilities move in fits and starts. God speed Wiseman, Glover, Koch, and Hansen! www.space.com/space-explor...
Wow, unexpected. Crew-11 returned to Earth early, January, due to medical emergency because Michael Fincke (549 days in space) had suddenly lost ability to speak. Trips to Mars are much more brutal in conditions and duration. May need better protection. Unknown. www.nytimes.com/2026/03/28/s...
Oh gosh. Remember the excitement about frontier models understanding medical images and acing tests? "In the most extreme case, our model achieved the top rank on a standard chest X-ray question-answering benchmark without access to any images." Fei-Fei Li is an author. arxiv.org/abs/2603.21687
Waymo doesn't have enough vehicles, prob bc
- they remain too unreliable to grow as fast as planned, ie overestimated their ODD expansion
- 6th gen aren't approved in California
They've been stuck at same growth rate for ~2 years & may miss EOY million ride/week target.
bsky.app/profile/anic...
My inference. Waymo doesn't have enough vehicles in SF. Last few days, Waymo surge pricing is way above Uber. Uber surge attracts more drivers to jump in, bringing surge prices back down. Waymo has fixed number of vehicles to deploy. They need to get more soon in SF. Demand is there. Tipping point?
Mustafa Suleyman chief of MicrosoftAI in Nature 19Mar2026 p559 "Humans have evolved to imagine the possibility of agency everywhere. When a system perfectly mimics intentionality and empathy, the human brain projects inner life into it...Seemingly conscious AI weaponizes this biological instinct."👍
My full time day job, CTO @2. Robust.ai
For building robots that help warehouse workers, not replace them
... among that growing group, Robust.ai stands out for its emphasis on developing situationally aware robots that can work alongside people. www.fastcompany.com/91497289/rob...
OTD saw excitement about using LLMs to break down tasks & tell individual robots what to do. Spoiler: teams of robots already use planners & optimizers for this. Telling a robot what to do is not the hard part. Having a robot in a messy real physical world succeed at doing what it is told to do is.
My AI can pass the bar exam but can't tell when a client is lying to it. One of those matters in court. It's not the exam.
VCs put $56B into defense startups in 2025, which has so far generated just $4.3B in contracts. None of these new weapon systems are yet at the test phase.
www.theinformation.com/articles/war...
4/4 I think most human observers do not understand this critical difference between modern robot training methods and human capabilities, leading them to grossly underestimate the time and cost to scale early demonstrations.
3/4 My inference is that Waymo's learning methods attach locations, rather than situations, to states, which then learn what to do for that state. Situations are much richer semantic entities than locations. Humans are great at both. Waymo learning bypasses situations but at a generalization cost.
2/4 These different operating modes of two Waymo vehicles in the same geography indicate a weakness in end to end learning compared to human driving. Once we know an area driving in one vehicle, we can tune up our driving of a new vehicle anywhere, then easily drive all over our known territory.
1/4 In the last week both driverful Zoox and Waymo Ojais (small blue vehicle based on Zeekr RT) have shown up en masse in Pacific Heights, Cow Hollow, and the Marina. Neither were in this part of SF before. Already, uncrewed Waymo Jaguar taxis operate continuously in these areas of the city.
Star Trek TOS episode 2:14 Spock tied up the computer (it emitted smoke) computing the last digit of pi. TNG had people asking the computer to generate large scale computations. Ridiculous. Versions of that now happening, but users need to understand and convey much more than depicted. New tools.
Actual risks from modern AI. $1.5T in existing investment grade bond debt for AI data centers. Then there is training debt. And no cash flow positive products deployed. The more all in on new AI that a company is, the more risk. Failures will ripple through the economy. www.fool.com/investing/20...
I thought it helpful to plot out the annual positive cash flows for all (every one of them) companies that have achieved that from deploying learning based robots (VLA, VLM, LLM, RL, and even more conventional DL for driving) this century, including AVs. Here they all are, named and plotted.
Data centers are now everywhere and part of critical infrastructure. They are now targets in wars. Amazon reports that two of its data centers were hit in the UAE and one in Bahrain overnight. Having visited data centers recently I suspect repairs will be complicated. www.cnbc.com/2026/03/02/a...
Changes to the NASA crewed lunar landing plans were inevitable. Level headed approach from new NASA administrator. This makes for more but smaller steps, opens things to both Blue Origin and SpaceX and rescues SpaceX from its own unrealistic hubris. www.nbcnews.com/science/spac...
I think it is a bubble. I got a copy: Brooks thinks the current wave of massive investments has probably gone too far. “I don’t think they’ll pay off the way some people have imagined,” he said. “I’m not an investment advisor. But yeah, I think there’s a danger.”
I think we just saw exactly that with LLMs. Their language proficiency is way beyond what we had before; their performance vs competence (one of my "7 sins") is making some people think we have reached human level intelligence. They have not. But it is a big software only jump. So, yes, possible.
