xpct

I'm quite fortunate to keep piling on knowledge without extrinsic rewards: the acquisition itself and the rewiring of my brain is what makes it enjoyable for me.

I look up to people who are well-read and try to follow their example. Maybe one day I can inspire somebody else.

My response was directed to your insincere characterization of Anthropic's actions. As we can see from the comments here, the public opinion hasn't settled the same way as the court case has, and that's why it's still discussed.

In my experience, there haven't been many cases where I can't assign a core or multiple of them to spinlocks, if it's in my hot path.

That said, I probably wouldn't ship spinlocks in consumer libraries or code I expect to be reused across deployments.

Interesting, I haven't heard this claim before. I suppose that claim made sense if their customers were big corporations, not so much when its the masses generating bootleg software copies.

Let's not sane-wash Anthropic's book theft. No, they didn't just 'scan' the internet, they created a tool for worldwide license washing and got fined an insignificant amount for it.

Is it so wrong to be doubtful of whether the field is still lucrative? We live in times of great change, it's only natural to be less assured when recommending a career path to others.

I find the operator point of view equally horrifying.

Likely, they will run an OCR scan to blur any text seen through the cameras. Also likely, they would run a vision model classifying identifiable information: people walking past, hanged photos or even paintings, TV screens, maybe even the view outside the window? For safety purposes, I imagine it's also worth identifying handheld tools such as knives, bats, etc., all done so the operator risks are minimized.

Fair enough, and I'm not arguing against that, there's just something distasteful about the operator having to walk around in a largely obscured, muted camera feed, looking for traces of underwear scattered around the house.

So, the idea itself is fine. The timing at which it's introduced is what makes me nauseous. We're really trying to milk the agents in any way possible, aren't we?

5 years ago, we may have expected that robots would write APIs that utilized every HTTP code, what we got was a json with a stringified error field..

Yes, that's the path we're on. It may start with poor eastern Europeans, then gradually move to Africans who tele-operate on eastern European homes.

I'm not just triggered by the wording, I generally just find it hard to read. That familiar feeling of having read multiple paragraphs where nothing of substance was said, and you realize the text was either heavily edited or generated.

I'd love to read an article about the topic, but can't recommend this one if you expect high informational quality.

Once again, the text is riddled with LLM'isms. Is this the new norm nowadays? Looking at OP's submission history, it's evident that they are utilizing HN for SEO farming.

A much more valuable discussion would be centered around the company's own website, which contains the same information, and doesn't require an LLM mediator: https://www.weaverobotics.com/isaac-1

In this case I'm seeing 'micromanaging' as something akin to automating a Tesla superfactory: machines complete each individual step, we get a working car at the end.

Obviously, we'd have processes built that build machines for each of the individual tasks, or changing tracks to support new car models.

From my perspective, this framework of a factory could map to many other endeavours where we either produce the end product or a machine used for something else.

So, the difference would be whether the machines utilize chemical/biological processes for working, or are made out of steel, at which point it would boil down to economics.

I guess I don't see what's so special about adding 'bio' to these perpetual factories.

Interesting thought experiment, but I don't see why automating machines that build and repair other machines wouldn't be sufficient. At the limit, such a machine would be able to repair itself, or repair other long-running machines. I imagine it would come down to wear and efficiency loss.

Yes, phasing out physical discs is predatory. I'd like to also add that buying a console which can only run vetted games has already been predatory, and digital games are only the next natural step.

So, you can't use it for coding, can't use it for 'sensitive' information in chemistry/biology. It follows that it's likely bad for medicine and adjacent topics too.

What can you use it for? To run a breadth search on Erdos problems?

I don't think OP argued for any type of relentless work.

> checking out emotionally and spiritually just means more life lived

What exactly about checking out makes it 'more living'? I sense a false dichotomy here

Has a16z already boasted about this investment, or is that podcast episode coming out later this year?

I wasn't working with anything sensitive, but it really does feel like it sometimes condenses even something low like three bullet points to two.

IMO, they were quite good with checklists even a year ago, and tried to tick off each one.

> Generally speaking, we know how the level of abstraction that we were born with works

I'd say it's generally true that the majority of jobs of an era deal with a similar level of abstraction, and that's why most people stay on it. However, I frame this as being born with technical debt, and it's my obligation as an engineer to understand what the previous generations have built, and where it makes sense for me to work, directionally.

> audit all the data and code that is associated

For a while now there has been very little incentive for providing these alongside the paper, and I don't see why exactly 'AI' would change this. I could even see how making it vague to be harder to test with LLMs could be profitable for citation hackers.

I'm actually quite excited for when (if) the models get good enough to start replicating compsci papers. I'd love it if there was a system which calculated a reproducibility score per-lab or per-researcher, which I could look up alongside their citation count.

I want to see who did the hard work properly, and who focused on publishing with concealed details.

I've been largely disappointed how much the Claude models ignore custom instructions, and sometimes even prompts on the chat interface. It sometimes feels like talking to a wall, or as if there was a third person in the chatroom whose messages I can't see.

I can't help but feel this is intentional towards the 'Agentic' workflow.

Thank you for sharing the link. It's fascinating that the models can only do 10-20% on the hard subset, and I wonder why that is so. The fact that they can only get 30-40% out of the fp8 GEMM seems unintuitive to me, I would've expected a convergence near ~80%.

To my knowledge there's still talent left in London, even after the wage stagnation.

I wouldn't be as sure that engineers won't be needed, whereas welders will.

I admire your persistence and hope your efforts pay off!

Agreed, I haven't yet internalized them not remembering feedback from a couple days ago, which humans typically would. The process of remembering to update the context isn't entirely natural to me to this day.

Would you fault those who've had outstanding 20+ year programming careers for not anticipating the change?