Mauricio Ordóñez

Machine learning at Amazon Core AI. Ex-Microsoft. Statements are my own opinions.

[Mauricio Ordóñez @mauordon]

@giffmana @elonmusk Maybe they neglected to jit the JAX training function.

[Mauricio Ordóñez @mauordon]

@alialagheband @a16z Humans also don’t require hundreds of millions of GPU compute hours and gigawatts of power to learn to drive.

[Mauricio Ordóñez @mauordon]

@Polymarket Pete Rose would like to have a word.

[Mauricio Ordóñez @mauordon]

@teslayoda Could this have something to do with it?

[Fox News @FoxNews]

'TERRIFYING': Dashcam video shows the moment a Tesla Cybertruck, allegedly operating in self-driving mode, nearly sent a Houston mom and her infant off a bridge before violently crashing into an overpass barrier.

The woman claims she suffered multiple injuries from the incident and is now suing the automaker for $1 million.

[Mauricio Ordóñez @mauordon]

@mercoglianos Can they go through Suez or do they have to go through Beb-el-Mandeb?

[Mauricio Ordóñez @mauordon]

@Adam_and_EVs How can we be at V4 when there is even a working minimum viable product?

[Mauricio Ordóñez @mauordon]

@trengriffin @leonpunk Radiators might not be new, but at this scale? A cluster of 72 H100s (assuming inference workloads and not training) might be around 40 kW. Consider the size and mass of the solar array and radiator. Are the unit economics better than terrestrial solar+batteries or nuclear?

[Mauricio Ordóñez @mauordon]

@nntaleb @Joshuaerom21861 I believe there’s a rolling resistance advantage to 38mm. The loss would come from aerodynamic drag from the wider cross section, which probably wouldn’t matter unless above ~30 kph.

[Mauricio Ordóñez @mauordon]

@Anthony_Bonato And epsilon = 1 / phi

[Mauricio Ordóñez @mauordon]

@RealJimChanos I’d also like to see the fixed costs for the cooling system required to dissipate all that heat using only radiation.

[spark @sparkjsdev]

Open Sourcing Forge: 3D Gaussian splat rendering for web developers!

3DGS has become a dominant paradigm for differentiable rendering, combining high visual quality and real-time rendering. However, support for splatting on the web still lags behind its adoption in AI.

[Mauricio Ordóñez @mauordon]

@mattturck Understandable, but also amazing to see Tensorflow left out of the slide.

[John Carmack @ID_AA_Carmack]

There is an alternate reality where Cray took their vector supercomputers, ditched FP64 calculations, and went with one FP32 pipe and a BF16 tensor core pipe. The same instruction set, memory architecture, and vector registers would have made a sweet deep learning machine, in many ways nicer than SIMT CUDA programming on GPUs. A Y-MP class machine like that could have delivered the AlexNet and DQN moments two decades earlier.

Even doing everything in FP64 with no architectural changes, a Cray-1 would have been the best machine in the world for neural networks. If @geoffreyhinton had access to one for early research, the case could have been made for the architectural modifications to 10x the performance.

[Mauricio Ordóñez @mauordon]

@trengriffin Is there an explanation to why this happened?

[Mauricio Ordóñez @mauordon]

@nicknotned That’s true about the hardware. However, much of the value of Apple products is in the software and services.

[Mauricio Ordóñez @mauordon]

@Riazi_Cafe_en While I liked Strang very much, Axler is the best.

[Nassim Nicholas Taleb @nntaleb]

Either I made an arithmetic mistake, or the risk factor from cycling is ridiculously small!
Compare to Rf for any other activity.

[Mauricio Ordóñez @mauordon]

@TheStalwart CDOs?

[Mauricio Ordóñez @mauordon]

@goodfellow_ian I’m sorry you are going through this.

[Information is Beautiful @infobeautiful]

Child's play
(creator unknown)