Hacker Newsnew | past | comments | ask | show | jobs | submit | dminik's commentslogin

I don't. I was recently doing some searching for information I thought AI would be good for: fuzzy natural language search with some conditions. And it was, but ...

Gemini at least is not great at citing and picking sources. Or providing multiple sources for the same thing.

It tends to stop at threes. So if you want more, you have to prompt it uselessly, like: "any more?"


Assuming that there are infinite suckers with cash to spend. It's entirely possible (if unlikely) that the market is not big enough to cover the training costs. Especially for multiple companies all burning insane amount of money on the regular.

I'm in these stats I think. But mostly because I was trying to do an exact search ("something to search") and discovered that google just ignores it.

There's a local search engine with a motto that translates to something like "Find what you don't know." Google has seemingly adapted "find what you don't want."


Ok, I have to ask. How do you accidentally turn it off? The power button is at the top and it's flush. It's not like you can hit it accidentally.

I hit it accidentally all the time. Any time I have to temporarily one-hand it with my right hand. The sides are rounded and smooth and there are buttons on the back and top of the side. The power button is very sensitive as well and doesn't require a hard press; it responds instantly.

Are we talking about the right button?

https://cdn.shopify.com/s/files/1/0253/3664/3635/files/641_e...

You would have to go out of your way to grip it in a way you could press it. You don't need to move your hand to lift it, it's controller shape after all. Or you can grip it along the bottom edge. And even gripping the top edge I just can't find a way you could accidentally hit it. It's flush.

The only time I've accidentally turned it on/off is when I've been clawing it out of the carrying case.

Edit: Wait, are you gripping the bottom and top edge at the same time, over the screen? Why? It's huge.


Yes, top and bottom at the same time, from the back or front. I have giant hands and that's a comfortable grip for me. Anything else feels like it's going to slip and drop one-handed.

> As a result, many of our Webflow teammates are leaving the company today

What a disgusting way to phrase that.


Why? How would you as a leader would announce this in a way that better reduces bad feelings?

Alternatively, can you explain why you believe this statement reduces bad feelings? Put yourself in the shoes of those who have been laid off.

Defend your position, I’m tired of, and will not play defense.


Well, first of all, stop lying. These people aren't leaving the company. They were fired. They were removed. Leaving sounds voluntary.

Then, this might be a cultural thing, but I don't want niceties and flowery language. Give it to me straight. It's not "we're rebuilding". We're not rebuilding anything. We're broke. If you're broke just say that.


> How would you as a leader would announce this in a way that better reduces bad feelings?

Step 1, stop using AI as an excuse?


Step 2, don't use "leaving the company" as a euphemism for "getting laid off", as the former implies employee agency and choice they don't actually have.

The entire post is a green-beige sludge of corpspeak, euphemisms, and AI slop. Who fucking talks like this anyway?


Direct and to the point. This is the opposite of that.

That's on me; I set the bar too low. When I asked "How", I was hoping for someone to literally write up an alternative and better way of phrasing this, which would still work as part of a corporate press release.

Asking this sort of question is not great in professional context either.

Someone working for the police could say: "Yeah, my boss made me clean up a triple homicide."

Or a janitor at a fast food could say: "We found a dead addict in the toilets."

Like these are all profession related answers. Yet they are not answers you want. Stop asking dumb questions.


Morbid curiosity is a thing, even if professional setting. I only know one person who got this kinds of questions when they applied for forensic technician jobs, collecting remains of dead bodies and such.

I don't really like it either.

I much prefer asking about things the candidate would take from or actively avoid from the previous job.


Kind of, yeah. LLMs are hyperoptimized for one-shotting. But all of their code is like that. It's poison for the long term health of the codebase.

Does Anthropic really expect to double their income without also doubling their expenses?

There we go back to the original question: are subscriptions profitable, API pricing wildly profitable, and they just lose all that money on fixed costs like model training; or do they actually barely make money on inference?

That's why talking about the profitability of inference without accounting for model training is interesting, because that is the deciding factor in whether more customers would help getting them in the green


Without actual data I don't know. My gut feeling is that they overall lose money on subscriptions (and especially the free tier that accounts for 95% of all users). And make thin profit (~5%) on API pricing.

But it's just that. A gut feeling.


I don’t think it’s a gut feeling. It seems to be the consensus that subscriptions are still heavily subsidized

its well reported that inference margins on api pricing are 30-50%

No it's not. It's well speculated. Can you give a single source that says they have a 50% margin on inference?

I’ve been hearing that anthropic is on the verge of profitability for probably a year straight. Until all the companies agree to stop the training arms race I just don’t see how it’s in the cards

This is one of the things people miss. If they double their customers, of course they double their expenses. Unlike SW, the marginal cost here is still high

I mean, it's possible that with the new datacenter from SpaceX, they could onboard more users than it costs them to rent. That's fair. But I kind of doubt that.

One thing that really stinks to me is that various AI boosters have been claiming insane profit margins (40%, 50%, ...), yet apparently Anthropic stands to (possibly) make $500M profit on $11B in expenses, that's clearly nowhere near 50%. Not to mention that they're not making profit on inference now.

So where do people get this confidence to pull random numbers from?


> Goroutines are a far more tractable primitive for coding agents than threads

Goroutines are literally threads. Yeah, this really is a "go is my fav" article.


I use both go and python in my day job. The code that Sonnet produces for Go is much better than the Python it creates.

This could be because our go code is typically smaller more defined services but I don't really believe that since even the isolated python services are pretty spaghetti looking.


Goroutines are not directly equivalent to threads.

If 100 goroutines are handled by 10 threads, the effect on correctness is identical: any two can be running in parallel with each other (not just concurrently). From the point of view of this discussion, that's all that matters.

Correctness is nice but performance characteristics exist too.

They used to not be, because they were cooperatively scheduled and threads can be preempted. But they added goroutine preemption in Go 1.14 so in practice there aren't really any significant differences to threads, at least in semantics. (At least as far as I remember; been a while since I wrote any Go.)

You can be pedantic and say they aren't technically threads but that doesn't really matter from a programming perspective.


Even when goroutines were cooperatively scheduled, because the cooperation was mostly hidden and every function call was a yield point the average developer would treat them as being cooperative... until they spawned too many goroutines with a tight loop (and no function call) and the runtime locked up.

> You can be pedantic and say they aren't technically threads but that doesn't really matter from a programming perspective.

They are technically threads: they are independently scheduled, concurrent units of execution sharing an address space. They're just not OS (or kernel) threads. Hell, technically userspace threads (generally cooperatively scheduled) are the original, they predate kernel threads by a decade or two.


As someone below said, they might be from programming perspective, but technically they are not. See GOMAXPROCS for more info.

That being said the whole `tractable primitive` thing used in the article sounds somewhat sloppy to me. I don't quite get it. Yeah, they could be easier for an agent to write than async/await, but threads are also trivial in that matter, and you'd still need a mutex with go routines.


Amazing. People here would rather be mad at fake racism than at real racism. I wonder if that correlates with how much they're invested in LLMs.

I'd personally describe the seemingly inevitable placement of toxic byproduct industries in poor minority areas as economic bigotry, I'll concede that in the USofA that veers more toward black neighbourhoods in some states and accept that it also includes a lot of poor white areas also.

Presumably no one said “let’s build the data center where the black people live, mwahahah.” Instead, they said, “let’s build the data center where it’s cheap.” Where it’s cheap also happens to be where the black people live, and that’s a pattern. The pattern has historical roots, but it also affects modern communities. The academic quoted in the article is likely highlighting this point, which probably to them seems obvious, given they are steeped in the field.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: