josaka · HackerTrans

w zeszłym rokudiscuss

I suspect it's, in part, because humidity is low. The fab needs to tightly control humidity and temperature. Pulling water from air is energy intensive. Facilities team in the fab I used to work in said their most energy intensive days where the most humid days.

josaka

3 lata temudiscuss

My intuition matches yours. It's abundant enough to find its way into alloys for jet engine blades at single digit percentages: https://www.thenakedscientists.com/articles/science-features... And if used in chips, you don't need to make the full substrate with this material, just a few hundred nanometers at the top, in the active area.

josaka

3 lata temudiscuss

Same fab, same experience. I would add that a deep understanding (and ability to explain) statistics was more critical than understanding the underlying device physics for most roles I encountered. Enormous amounts of noisy data, characterizing several hundred process steps, is harnessed to apply massive pressure on engineers to "fix" problems that might just be a statistical blip. Economic stakes of minor yield deviations are so high that you often have to act as if something might be wrong before signals reach statistical significance. I bailed after four years, but it was great experience to start my career.

josaka

3 lata temudiscuss

Lawfare did a nice analysis of whether Section 230 shields model hosts (likely no): https://www.lawfareblog.com/section-230-wont-protect-chatgpt Professor Volohk has a public draft article on liability for output of AI models, with an emphasis on defamation: https://www2.law.ucla.edu/volokh/ailibel.pdf

Both suggest that this is a real area of concern. Folks have gotten complacent that Section 230 protects platform providers, but it is not clear that the protection extends to outputs from generative models, indeed one of the Supreme Court Justices implied that it would not in oral arguments for one of the recent Section 230 cases.

josaka

3 lata temudiscuss

It would be wonderful if it was easy for most people to represent themselves. My sense is most people struggle to express themselves precisely and clearly enough to avoid being at a massive disadvantage to someone who can do so. To my untrained eye though, AI seems poised to get us there. What LLMs are doing for code, translating imprecise natural language expressions of intent into machine-readable, precise code, looks similar to what I do as a lawyer when translating between a client request and legal work product. Lots of filling in boilerplate defaults, some assumptions based on context, etc. Differences include that sometimes we engineer ambiguity into that work product, and we can assume an adversary will exploit plausible interpretations unfavorable to our side, but I see no reason AI won't be able to do that as well.

comments