- this is not correct
- Is that the strawman framing of my argument that you want to stick with?
https://www.defensenews.com/global/europe/2024/11/11/why-swe...
- It's interesting that people are very incredulous of there being a legitimate defense reason for this when we have had unilateral unanswered drone incursions all over Europe and the US.
There's obviously some sort of arms race occurring and some of it is public.
The world in on the precipice of many technologies advancing at an all too rapid pace. The idea that technology will become tightly regulated isn't inconceivable.
FYI Sweden did the same thing last year. There is likely a (drone) reason, it's all but completely clear.
- > It was built by a team of former engineers from Dutch semiconductor giant ASML who reverse-engineered the company’s extreme ultraviolet lithography machines (EUVs)
This seems like the obvious conclusion of an ethnic bloc against a mercenary creedel nation?
Any westerner reading this right now wouldn’t die for their country, it’s almost absurd. It’s like asking them to die for Walmart.
- The models aren't smart enough to be fully agentic. This is why Claude Code human-in-the-loop process is 100x more ergonomic.
- I'm going to be honest because I have nothing to lose; I've never in my entire life met a single senior engineer that can surpass Opus generally. Some can surpass it at problem solving but they aren't common.
- We know what it's good at - software engineering.
- Correct. Opus 4.5 'solved' software engineering. What more do I need? Businesses need uncapped intelligence, and that is a very high bar. Individuals often don't.
- How many times have the goal-posts shifted now?
Everyone is justifiably afraid of AI because it's pretty obvious that Claude Opus 4.5 level agents replace developers.
- Nano Banana Pro is so good that any other attempt feels 1-2 generations behind.
- I think you’re saying this with unwarranted confidence.
- Yes, they're purposely not 'trained on' chain-of-thought to avoid making it useless for interpretability. As a result, some can find it epistemically shocking if you tell them you can see their chain-of-thought. More recent models are clever enough to know you can see their chain-of-thought implicitly without training.
- That does feel a little more like over-fitting, but you might be able to argue that there's some philosophical proximity to lying.
I think, largely, the
pipeline would obviously produce 'lying'. The trainings are in a sort of mutual dissonance.Pre-training -> Post-training -> Safety/Alignment training - Claude's version:
As usual, a non-marginally superior mind to commentators.even, odd = "even", "odd" for i in range(2\*32): print(f' if (number == {i}) puts("{even}");') even, odd = odd, even - They really lie.
Not on purpose; because they are trained on rewards that favor lying as a strategy.
Othello-GPT is a good example to understand this. Without explicit training, but on the task of 'predicting moves on an Othello board', Othello-GPT spontaneously developed the strategy of 'simulate the entire board internally'. Lying is a similar emergent, very effective strategy for reward.
- Yes, but it's not good enough. They needed to surpass Opus 4.5.
- The irony is that Deepseek is still running with a distilled 4o model.
- Professionals at anything, let alone an elite performance sport like this, are almost certainly statistically more intelligent than average.
Would would we want to purposely decrease interpretability?
Very strange.