4 ms·
> That's full multi-modal training with embodied agents (aka robots). 1x, Figure, Physical Intelligence, Tesla are all making rapid progress on functionality wh
by sincerecook 2y ago
> That's full multi-modal training with embodied agents (aka robots). 1x, Figure, Physical Intelligence, Tesla are all making rapid progress on functionality which is definitely beyond frontier LLMs because it is distinctly different.
Cool, but we already have robots doing this in 2d space (aka self driving cars) that struggle not to kill people. How is adding a third dimension going to help? People are just refusing to accept the fact that machine learning is not intelligence.
- warkdarrior 2y ago> Cool, but we already have robots doing this in 2d space (aka self driving cars) that struggle not to kill people. How is adding a third dimension going to help? If we have robots that operate in 3D, they'll be able to kill you not only from behind or from the side, but also from above. So that's progress!
- akomtu 2y agoMy understanding is that machine learning today is a lot like interpolation of examples in the dataset. The breakthrough of LLMs is due to the idea that interpolation in a 1024-dimensional space works much better than in a 2d space, if we naively interpolated English letters. All the modern transformers stuff is basically an advanced interpolation method that uses a large local neighborhood than just few nearest examples. It's like the Lanczos interpolation kernel, using a 1d analogy. Increasing the size of the kernel won't bring any gains, because the current kernel already nearly perfectly approximates an ideal interpolation (a full dataset DFT). However interpolation isn't reasoning. If we want to understand the motion of planets, we would start with a dataset of (x, y, z, t) coordinates and try to derive the law of motion. Imagine if someone simply interpolated the dataset and presented the law of gravity as an array of million coefficients (aka weights)? Our minds have to work with a very small operating memory that can hardly fit 10 coefficients. This constraint forces us to develop intelligence that compacts the entire dataset into one small differential equation. Btw, English grammar is the differential equation of English in a lot of ways: it tells what the local rules are of valid trajectories of words that we call sentences.
- tick_tock_tick 2y agoI ride in self driving cares basically once a week in SF (Waymo). It's always felt safer then a Uber and makes ways less risky maneuvers.
- n_ary 2y agoCould be because Uber or Taxi is trying to make most trips and maximize day earning while Waymo do not have that rush and can take things slow… Of course Waymo needs money but if the car made fewer trips compared to Uber/Taxi, it is not suffering the same consequences. We need to consider human factor and the severe lacking of that in these robot/self driving/LLM and drawing parallels is not a direction I am feeling comfortable. End of the day, Tesla also sold half baked self drive that killed people, we should not forget.
- soheil 2y agoHow is self-driving a 2D problem when you navigate a 3D world? (please do visit hilly San Francisco sometime) not to mention additional dimensions like depth, velocity vectors among others.
- physicsguy 2y agoThe visual input and sensory input to the self driving function are of the 3D world but the car is still constrained to move along a 2D topological surface, it’s not moving up and down other than by following the curvature of that
- soheil 2y agoSo based on your argument they actually operate in 1D since roads go in one direction and lanes and intersections are constrained to a predetermined curly line.
- physicsguy 2y agoThe point is clearly that they don’t have a vertical axis of control, they can’t make the car fly up in the air unless they’re driving crazy taxi style