Massachusetts Institute of Technology, a leading research university with prominent AI safety researchers.
“MIT's Andrew Sutherland frames this as a qualitative leap”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.
“Per-step action-decoding latency drops by up to 20× (Section 1, Section 3.1)”
Source→“FlashVLA is a drop-in modification to any flow-matching-based VLA model, requiring only light architectural changes and a fine-tuning pass”
Source→“With low trust, every equilibrium races to ruin: the disaster arrives with probability one. With intermediate trust, immediate stopping and racing to ruin are both equilibria.”
“Most legged robot research focuses on small, highly dynamic platforms (like the MIT Mini Cheetah) that can recover from failures using rapid, aggressive stepping.”
Source→“95% of generative AI pilots produced no measurable business impact, according to MIT's Project NANDA.”
Source→“Max started promoting AI for Physics from 2012 — that was an extremely early, extremely forward-looking move. It's because he has very broad interests and can see connections between different things.”
Source→“A new wave of "New Labs" is emerging in both the US and China, founded by young researchers with elite academic backgrounds who can raise hundreds of millions in valuation before having a product. Liu Ziming describes the fundraising environment as almost irrational in its speed and intensity.”
Source→“Anthropic's work, including sparse autoencoders and transcoders, is all based on this idea — being able to extract atomic concepts from superimposed internal representations. Members of the interpretability team like Samuel Marks and Wesker Nee — and an undergraduate I mentored at MIT — all ended up at Anthropic.”
Source→“Max threw a 1989 paper at me by Thomas Poggio, who is a very famous professor at MIT. That paper argued that the Kolmogorov-Arnold theorem could not be turned into a real algorithm.”
Source→“MIT's Project NANDA studied 300 public AI deployments and found that "95% of generative AI pilots failed to produce measurable financial return," with failures traced to "poor integration and mismatched priorities, not weak models."”
Source→“Pioneer of data abstraction (CLU language), distributed systems (Argus), ViewStamp Replication, the Liskov Substitution Principle, and Practical Byzantine Fault Tolerance (PBFT).”
Source→“For view stamp replication, it was a delay of about 10 years before people started to use it.”
Source→“Gil Pratt: Former MIT professor, later at Toyota Research Institute. Inventor of series elastic actuators — the insertion of a physical spring into a robotic joint to eliminate reflected inertia at high frequencies.”
Source→“One thing we haven't done is to make robots trained with machine learning be compliant. We came up with this new approach that we call soft mimic.”
Source→“If you make the machine look like a collection of springs and masses and viscous elements, that was sufficient to make sure that it was stable in contact with any object.”
Source→“Our robots not only follow the motions that we command them to be, but also they are smart about how much force they exert. For example, if the robot is carrying a cup of tea but it hits a table, the robot will adjust and absorb the shock.”
Source→“A robot that was stable during free motion could become unstable when you asked it to push on a surface... if you make the machine look like a collection of springs and masses and viscous elements, that was sufficient to make sure that it was stable in contact with any object.”
Source→“Learning trained robots usually understand where they should move their hands and legs. They have trained to mimic a wide variety of motions like dancing in many many different ways but they're so focused on replicating that motion that they don't account for how much force they're applying. It can damage the object, injure me or break itself.”
Source→“Our lab has pursued this approach of training robots in a simulation. The digital replica of the physical world, where robots can be collecting data in a digital equivalent of reality. In just a few hours, we can collect hundreds of days' worth of data.”
Source→“MIT researchers in an April paper described AI automation as more of a "rising tide" than a "crashing wave."”
Source→“"In the last five months, I've seen the emergence of what I like to joke with my wife as the Bernie to Bannon coalition... Extremely unlikely bedfellows from across the whole political spectrum saying, 'This is crazy. This is absolutely nuts. Let's do something about it.'" — Max Tegmark, MIT”
Source→