A sequence of events:

  - Democrats win congress : it sets a clock to act before January
  - Jensen is taken out, throwing nVidia into chaos
  - Email comms are blacked out for an entire day across government and major enterprises
  - Air traffic is stopped nation wide the next day

Because while a founder CEO can play fast and loose with gray finance rules, I don't expect an appointed CFO can do the same. At least not one that expects to work again if they are found liable for it. It is kind of interesting in this way that the board has made this choice. It almost says they are looking to insulate against some liability here.

They can't legally not fire him. But they can legally re-hire him after firing him. So ironically the ONLY way he can continue at nVidia is by being fired - we just have to wait a few days and there will be an Altman-like resurrection.

This is the right line of thinking, but I don't think a second affair is enough. To me, it says there is more we don't know. Possibly a lot more, and a lot worse than a second affair. Apart from anything else, that is the only reason to act this fast. You don't fire a CEO like Jensen in 24 hours over an HR indiscretion.

So I think both you and the parent are correct, but its a combo: he didn't come fully clean, but there is a non-public part that probably looks a lot more like treason that the board knows and we don't.

I figured there will be quite a lot of people taking actions like this ... but nVidia was a surprise for sure.

It would only be part of the answer but the most striking thing to me is that AI lacks the inherent sense of minimalism most humans have because creating code is so hard for them. It's so hard that a large part of our initial thinking will simply be to minimise the solution down to the least possible change from current state.

AI tools do have a little bit of this but I think it's heavily in conflict with more recent efforts to increase their persistence and the scope they can undertake and length they can work on something. All these have been optimised to hill climb coding benchmarks, but put against the benchmark of "are you sure you really need this at all? Couldn't we do it a simpler way? What if we just did it X way and then we only need to change 10 lines of code" - they are not even being tested against this metric.

The agent should not be running on the same workstation. It should have a hard isolation barrier between it and the host machine. We shouldn't be starting from a fail-unsafe state and then trying to patch in security and isolation for these people. It should be deployed fail-safe and when they hit actual friction we can solve that.

Anthropic started out doing the right thing with Cowork (running in a VM). But it seems like they've ever since been walking that back and trying to get users to rely on automatic protections. I think it's just the wrong direction.

The actual post seems incoherent. They had a bunch of problems with Claude, most in features Pi does not even support in the first place. They don't like "instability" aka rapid pace of development, ignoring Pi is just as if not more rapid. They passively assume an ecosystem of plugins is going to be more stable.

I'm a big fan of Pi and believe everyone should use it or similar open harnesses in preference to proprietary / vendor controlled ones. But we need to employ actual cogent, coherent arguments in advocating for that. When a broader audience tries open harnesses and finds out they don't live up to expectations the blow back will be costly.