francescovv

Hanlon's razor is useful to curb one's paranoia, but it is far from being a universal rule.

In fact, malice and incompetence are not necessarily mutually exclusive.

This very incident shows several instances where "Jia Tan" is being arguably incompetent, in addition to being clearly malicious: unintended breakage by adding extra space between "return" and "is_arch_extension_supported"; several redundant checks for `uname` == "Linux"; botched payload, so "test files" had to be replaced, with pretty fishy explanation; rather inefficient/slow GOT parsing, list goes on...

Wait, lens can affect colours?

That was between two Android phones, with receiving phone being "oppo" or some such, with stock firmware. Receiving phone would see incoming file request, ask user to accept, then error out with "unknown file", and no way to actually save it. I've sent files via bluetooth from lineageos to lineageos, no problem.

> What (...) happened to easily sending data, over the Internet

And for mobile phones - without internet is similar, unnecessarily hard. The other day I was hiking with friends, and wanted to share a .gpx file with the route, at some spot with no cell coverage. I thought: "I 'member, bluetooth can send files". Well, we spent good 15 minutes trying and miserably failed, that's no longer possible in the name of "security". So I had to wait for cell signal to come back and send the file via whatsapp. To someone standing right in front of me.

Alternative would be to have model itself aware of sensitive topics and and meaningfully engage with questions touching such topics with that awareness.

It might be a while till that is feasible, though. Until then, "content safeguards" will continue to feel like overreaching, artificial stonewalls scattered across otherwise kinda-consistent space.

> 100s of dependencies. If there is a 1% chance of a random repo having a backdoor, the project will be compromised

Apologies for nit-picking, but that's not quite how sum-of-probabilities work. Total probability across 200 tries of 1% chance each, is ~87%:

  p=0
  for _ in range(200):
    p=p+(1-p)*.01
  print(p)

  0.8660203251420382

Your "sooner or later, to the point where we can assume" conclusion, still stands, of course.