Last I tried, there's still often (not always) fighting with typing, mypy, pyright because of the dynamic nature you mention. I think the complaints about it are sometimes misinterpreted as "disliking types".
Interesting. So 516 probably means initial 512 byte buffer and a 4 byte header. Then 516 + 518 = 1034...so another 512 + 4 byte header + 2 bytes for a linked list ref or similar, 1034 + 518 = 1552, etc.
One area I think is really going to get slaughtered by LLMs are marketplace plugins. Those monthly fee plugins people release for things like Jira, Shopify, Salesforce, etc. There's a subset of those that don't have some backend that's hard to replicate, and asking an LLM to reverse engineer and make your own plugin is trivial.
The research here is clearly interesting, but if you just need to get something like this working, premade neon tube electrodes are plentiful and inexpensive.
Consider all of that can be used for forced confessions and forced plea bargains also. In those cases, the "evidence" doesn't even need to exist at all, or be on the record in any way.
Once the end-user fail-safe of "use a real heuristic ad blocker that's hard to get around" is gone...the incentive for ad platforms to get around the relatively easy hostname based blockers goes WAY up. They know it won't drive people to more sophisticated ad blocker if the technical barriers for those are high.
Google's playbook of slowly eeking out this stuff so that you don't notice you're in the boiling pot has played out several times.
No, I'm not. Yes, those companies exist. And, so do many companies on the other end. Where they bend over backwards to ensure their data only lands in places where they have the exact contractual language they want. Any stodgy F500 typically falls in that category. They would not likely be using Anthropic through the AWS "bridge" in the first place if they were chasing latest/greatest.