yo-scot-99

this tech is interesting but so poorly understood that it's just using the (public) roads as one large alpha test. given a NN there is no way to verify what safety ranges are there. for instance if each camera slightly changed exposure or occlusion are the results smoothly changing? all they can do is try it and hope the inputs are in a safe part of their optimization space.

also donating to reality-based candidates; even those who want to break up larger companies at least align themselves with a bit more with science and scepticism in things as they are. And to be honest as an employee if you end up in Facebook.codebase.1 or Facebook.codebase.2 who cares, your downside is minimal.

there's 200-300k cars out there with just cameras + radar and their bet is the software catches up to the rhetoric. Adding lidar is next to impossible on the current fleet and adds to the cost if they start now. It is an early design bet which the coders now have to meet.

A couple of meta questions for the HN community:

1. What percent of HN readers have a NYT subscription?

2. There are far more comments here than the source -- is that appropriate? Should the comments ride along w/ the article (for future reference and the fact that they paid for the orginal story/reporting)

I read it more as a "here's the keys to the kingdom" and referenced the redirectMethod which even has a step-by-step guide to replicate it. The author is from a search engine marketing consulting firm so it would seem to be almost against his self-interest to lay bare these tactics.

The Telsa autonomy investor day (which applied to the cars not the people) spent most of their time explaining their NN chip, ghost riding to correct predictions etc. Definitely gave the impression that the network was doing a lot of the decision making.

> Paul was always good at seeing the big picture.

This reminds me of a day at Asymetrix - a startup founded by Mr Allen. Somehow we got a very early incarnation of Mosaic and I remember showing everyone how to bounce over to each lab with just a click and what authoring in HTML meant. Asymetrix's main product at the time was Toolbook, essentially a Hypercard clone -- but this was obviously the future. Paul's reaction was muted but I think he knew "This is happening without us!"