catigula

12 ms·

241.
by catigula 9mo ago
If you can't make a simple chatbot safe, how does OpenAI square this with their open bid to build "superintelligence"? If the simple playmobile version is verifiably unsafe, why would the all-powerful god be safe?
242.
by catigula 9mo ago
People need to start having conversations about existential risk here. Hinton, Nobel Prize winner in AI, thinks there's a decent chance AI executes the entire human species. This isn't some crank idea.
243.
by catigula 9mo ago
Don't worry, AI will be nice for some reason.
244.
by catigula 9mo ago
AI isn't going to collapse. It's the real deal.
245.
by catigula 10mo ago
You're going to be eating so much crow shortly.
246.
by catigula 10mo ago
> but it's increasingly looking like LeCunn is right. This is an absolutely crazy statement vis-a-vis reality and the fact that it’s so upvoted is an indictment of the type of wishful thinking that has grown deep roots here.
247.
by catigula 10mo ago
The implicit subtext of 'WEIRD' is "these people are amazing and that's weird" tbh.
248.
by catigula 10mo ago
Yes. It's not just me: I'm a professional staff engineer, great. Andrej Karpathy is one of the best engineers in the country, George Hotz is one of the best engineers in the country, etc.
249.
by catigula 10mo ago
Nope. It's still faster to just prompt Claude and read all of the output, I'm sorry.
250.
by catigula 10mo ago
>build what they envisioned as a military-grade home security system for tech elites It must be feel so good lording over the plebs as a fiery eye.
251.
by catigula 10mo ago
Many developmental disorders can be 'cured' through compensatory approaches and probably gene therapies.
252.
by catigula 10mo ago
This seems remarkably less safe? Would would we want to purposely decrease interpretability? Very strange.
253.
by catigula 10mo ago
this is not correct
254.
by catigula 10mo ago
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...
255.
by catigula 10mo ago
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 occ
256.
by catigula 10mo ago
> 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 merc
257.
by catigula 10mo ago
The models aren't smart enough to be fully agentic. This is why Claude Code human-in-the-loop process is 100x more ergonomic.
258.
by catigula 10mo ago
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.
259.
by catigula 10mo ago
We know what it's good at - software engineering.
260.
by catigula 10mo ago
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.
261.
by catigula 10mo ago
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.
262.
by catigula 10mo ago
Nano Banana Pro is so good that any other attempt feels 1-2 generations behind.
263.
by catigula 10mo ago
I think you’re saying this with unwarranted confidence.
264.
by catigula 10mo ago
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 m
265.
by catigula 10mo ago
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 Pre-training -> Post-training -> Safety/Alignment training pipel
266.
by catigula 10mo ago
Claude's version: even, odd = "even", "odd" for i in range(2\*32): print(f' if (number == {i}) puts("{even}");') even, odd = odd, even As usual, a non-marginally superior
267.
by catigula 10mo ago
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&
268.
by catigula 10mo ago
Yes, but it's not good enough. They needed to surpass Opus 4.5.
269.
by catigula 10mo ago
The irony is that Deepseek is still running with a distilled 4o model.
270.
by catigula 10mo ago
Professionals at anything, let alone an elite performance sport like this, are almost certainly statistically more intelligent than average.

More ›