In the era of algorithmic distraction, the ability to maintain a single thread of thought for four hours is a superpower. It is the only way to solve hard problems.
It is becoming clearer that Jevons paradox applies to competent human software engineers. If AI makes them more efficient and more productive, demand for their work will increase.
Spend an hour reading this weekend and I think you’ll know more about robotics than 99% of people, including some people who invest in robotics.
https://t.co/eCqUAwXvgf
There will always be a place for special-purpose machines highly optimised to do one thing & one thing well.
You're not going to beat the speed and robustness of pneumatics chained together to replicate a series of simple movements.
Today, we're releasing ARC-AGI-2. It's an AI benchmark designed to measure general fluid intelligence, not memorized skills – a set of never-seen-before tasks that humans find easy, but current AI struggles with.
It keeps the same format as ARC-AGI-1, while significantly increasing the signal strength it provides about a system's actual fluid intelligence. Expect more novelty, less redundancy, and deeper levels of concept recombination. There's a lot more focus on probing abilities that are still missing from frontier reasoning systems, like on-the-fly symbol interpretation, multi-step compositional reasoning, and context-dependent rules.
ARC-AGI-2 is fully human-calibrated. We tested these tasks with 400 people in live sessions, and we only kept tasks that could reliably be solved by multiple people. Each eval set (public, private, semi-private) has the exact same human difficulty – average people in our test sample achieve 60% with no prior training, and a panel of 10 people achieve 100%.
Europe is home to great talent (5/10 top unis)
But we breed IP that mostly benefits others eg many drone components invented in Europe, almost all made in China now
We need a modern-day European DARPA to support deep tech and regional champions; tax credits won't make the cut
Redefining Tactile Sensing in Robotics!
We’re excited to unveil a new approach that transforms 6-axis force torque sensors into powerful tactile sensors, bringing robotics closer to human-level dexterity.
🔍 The Hidden Challenges of Tactile Sensors
Optical, camera-based tactile sensors have advanced robotic perception, but they come with limitations that impact real-world applications:
⚠ Complex Calibration – Requires frequent adjustments for accuracy
⚠ Light Sensitivity – External lighting can interfere with performance
⚠ Durability Concerns – Fragile optical components in demanding environments
⚠ High Computational Load – Processing large image data in real-time
⚠ Latency Issues – Slower response times in dynamic tasks
As robotics evolves, we need robust, scalable, and efficient tactile sensing solutions. Our approach leverages 6-axis force torque sensors to overcome these limitations—enabling precise force control, slip detection, and pressure mapping without the drawbacks of optical systems.
The future of robotic sensing is here! Let's work on it together🚀
Concorde was one of the most technically challenging projects of the 20th century: A commercial jet cruising at 2x the speed of sound. Nothing remotely close has flown since.
A symbol of what Europe can achieve when it dreams big and builds
The story of Concorde on @adastrashow
A Robotic blast in Athens!
So many talented engineers ready to drive innovation.
Many thanks to our special guests that shared their experience:
Dario Belicoso
@minasliarokapis
Eris Dhionis Sako
@IPatsiaouras@Klajd_Lika
…and many alumni from from NTU Athens that are driving innovation worldwide.
Many thanks to @BotaSystems for organizing.
Jeremy Berman, former #1 on the ARC-AGI-Pub leaderboard, just open-sourced his solution as a template on Params (link in next tweet)
It uses LLM-driven genetic programming, which has turned out to be way more powerful than anybody expected. You can book a consulting call with Jeremy via Params to talk about how the same approach could apply to your use case.
People have completely rewritten what "scaling laws" were supposed to mean. Originally it was "pretraining a larger LLM on more data leads to more intelligence" (which was confusing "intelligence" with memorized knowledge/skills)
Now it has somehow become "if we keep iterating on our models to refine their architecture, make them ever more sophisticated, and take advantage of ever more compute, we'll get better models". Which, duh...
Completely open source, including RL training, SimtoSim, and SimtoReal source code, ensuring that it can run on Unitree H1, H1-2, and G1 robot 😘
https://t.co/itp0m1Ifb4
A mighty robot manipulator… the rest are good but this is the best! 😜
Eris Dhionis Sako and the team from Duatic AG has been working on a semi-stealthy mode, but soon they will be showing off what the true potential of robotics is.