HNHacker News
TopNewBestAskShowJobs

noodletheworld

444 karma · joined December 14, 2024

submissionscomments
noodletheworld··on Are OpenAI and Anthropic losing money on inference?
Huh.

I feel oddly skeptical about this article; I can't specifically argue the numbers, since I have no idea, but... there are some decent open source models; they're not state of the art, but if inference is this cheap then why aren't there multiple API providers offering models at dirt cheap prices?

The only cheap-ass providers I've seen only run tiny models. Where's my cheap deepseek-R1?

Surely if its this cheap, and we're talking massive margins according to this, I should be able to get a cheap / run my own 600B param model.

Am I missing something?

It seems that reality (ie. the absence of people actually doing things this cheap) is the biggest critic of this set of calculations.

noodletheworld··on DeepSeek-v3.1
Security through obscurity is not security.

Your api key is linked to your credit card, which is linked to your identity.

…but hey, youre right.

Lets just trust them not to be cheating. Cool.

noodletheworld··on DeepSeek-v3.1
How can a benchmark be secret if you post it to an API to test a model on it?

"We totally promise that when we run your benchmark against our API we won't take the data from it and use to be better at your benchmark next time"

:P

If you want to do it properly you have to avoid any 3rd party hosted model when you test your benchmark, which means you can't have GPT5, claude, etc. on it; and none of the benchmarks want to be 'that guy' who doesn't have all the best models on it.

So no.

They're not secret.

noodletheworld··on Weaponizing image scaling against production AI systems
> Is it part of the multi-modal system without it being able to differenciate that text from the prompt?

Yes.

The point the parent is making is that if your model is trained to understand the content of an image, then that's what it does.

> And even if they can't, they should at least improve the pipeline so that any OCR feature should not automatically inject its result in the prompt, and tell user about it to ask for confirmation.

That's not what is happening.

The model is taking <image binary> as an input. There is no OCR. It is understanding the image, decoding the text in it and acting on it in a single step.

There is no place in the 1-step pipeline to prevent this.

...and sure, you can try to avoid it procedural way (eg. try to OCR an image and reject it before it hits the model if it has text in it), but then you're playing the prompt injection game... put the words in a QR code. Put them in french. Make it a sign. Dial the contrast up or down. Put it on a t-shirt.

It's very difficult to solve this.

> It's hard to believe they can't prevent this.

Believe it.

noodletheworld··on How to stop feeling lost in tech: the wafflehouse method
This isn't realistic.

Wanting things to be true does not make them true.

“Get a promotion this year, be a manager next year, manage the division in three years” is not a plan you can execute.

This is just the old self affirmation stuff you hear all the time: you won't succeed if you want it a bit. You wont succeed if want it and do nothing. You will succeed if you go all in, 100%.

It is BS.

You wont succeed if you go all in, statistically.

You might get a different outcome, but you wont hit your goal.

It is provably false that everyone who goes all succeeds; Not everyone gets to be an astronaut, no matter how hard they work.

The reality is that some people will put a little effort in and succeed, and some people will put a lot in and succeed. Other people will fail.

Your goals are not indicators of future success.

Only actual things that have actually happened are strong signals for future events.

The advice of having goals is helpful, but the much much more important thing to do is measure what actually happens and realistically create goals based on actual reality.

Try things. Measure things. Adopt things that work. Consciously record what you do, how it goes, how long it takes and use that to estimate achievable goals, instead of guessing randomly.

noodletheworld··on Show HN: Project management system for Claude Code
Are you talking about the same thing as the OP?

I mean, the parent even pointed out that it works for vibe coding and stuff you don't care about; ...but the 'You can't' refers to this question by the OP:

> I really need to approve every single edit and keep an eye on it at ALL TIMES, otherwise it goes haywire very very fast! How are people using auto-edits and these kind of higher-level abstraction?

No one I've spoken to is just sitting back writing tickets while agents do all the work. If it was that easy to be that successful, everyone would be doing it. Everyone would be talking about it.

To be absolutely clear, I'm not saying that you can't use agents to modify existing code. You can. I do; lots of people do. ...but that's using it like you see in all the demos and videos; at a code level, in an editor, while editing and working on the code yourself.

I'm specifically addressing the OPs question:

Can you use unsupervised agents, where you don't interact at a 'code' level, only at a high level abstraction level?

...and, I don't think you can. I don't believe anyone is doing this. I don't believe I've seen any real stories of people doing this successfully.

noodletheworld··on Claudia – Desktop companion for Claude code
Is a whole IDE really the solution though?

There are already a plugins to use claude code in other IDEs.

This “Ill write a whole IDE because you get the best UX” seems like its a bit of a fallacy.

There are lots of ways you could do that.

A standalone application is just convenient for your business/startup/cross sell/whatever.

noodletheworld··on Show HN: Yet another memory system for LLMs
? Are you complaining about MCP or boost?

It’s an optional component.

What do you want the OP to do?

MCP may not be strictly necessary but it’s straight in line with the intent of the library.

Are you going to take shots at llama.cpp for having an http server and a template library next?

Come on. This uses conan, it has a decent cmake file. The code is ok.

This is pretty good work. Dont be a dick. (Yeah, ill eat the down votes, it deserves to be said)

noodletheworld··on Ask HN: Do you struggle with flow state when using AI assisted coding tools?
When you stop and wait for an LLM it breaks flow.

If you stop and review what an agent did, it breaks flow.

Personally, my experience has been the best with doing this:

- Start coding

- On a second laptop, run an agent in yolo and tell it to periodically pull changes and if there are any `AGENT-TODO:` items in the code, do them.

- As you code, if you find something irritating or boring, `AGENT-TODO: ...` it.

- Periodically push your changes.

- Periodically pull any changes your agent has pushed down.

- Keep working; don't stop and check on the agent. Don't confirm every action. Just yolo.

If that's too scary, have it put up PRs instead of pushing to the live branch. /shrug

...but, tldr: if you're sitting there watching an agent do things, you're not in flow. If you're kicking multiple agents off and sitting waiting for them, you're more productive, but that is absolutely not flow state.

Anyone who thinks they are, doesn't know what flow state is.

The key to maintaining flow is having a second clone, or a second machine or something where you can keep doing work after you kick the agent off to do something.

(yeah, you don't need a second laptop, but it's nice; agents will often run things to check they work or steal ports or run tests that can screw with you if you're on the same machine)

noodletheworld··on GitHub CEO Warns Developers: "Either Embrace AI or Get Out of This Career"
> You know what else we noticed in the interviews? Developers rarely mentioned “time saved” as the core benefit of working in this new way with agents.

> They were all about increasing ambition. We believe that means that we should update how we talk about (and measure) success when using these tools

I’m struggling to understand what this means.

noodletheworld··on Deep Agents
This matches my expectations.

Now that its increasingly clear that writing MCP servers isn't a winning strategy, people need a new way to jump on the band wagon as easily as possible.

Writing your own agent like geminin and claude code is the new hotness right now.

- low barrier to entry (tick)

- does something reasonably useful (tick)

- doesnt require any deep ai knowledge or skill (tick)

- easy to hype (tick)

Its like “cursor but for X” but easier to ship.

Were going to see a tonne of coding agents built this way, but my intuition is, and what Ive seen so far, is theyre not actually introducing anything novel.

Maybe having a quick start like this is good, because it drops the value of an unambitious direct claude code clone to zero.

I like it.

noodletheworld··on I know when you're vibe coding
> The author may call it "vibe coding", that's fine but it has little to do with LLMs.

Humm.

Maybe if we say that this is not an issue from vibe coding it wont be?

Maybe if we pretend that maybe a naive junior would make these mistakes (true) we should be happy to accept them from senior developers (false)?

LLMs are extraordinarily bad at doing these things.

I’ve seen it.

You've seen it.

The OP has seen it.

You’re in a rush so you wrote some classes in a code base in a language which supports classes but has no classes in it?

Really? Did that get past code review before? Did you deliberately put up a code review that you knew would be rejected and take longer to merge as a result because you were in a hurry?

Of course not.

You did the bare minimum that still met the basic quality standards expected of you.

I get it. We all get it. When youre in a rush you cut corners to move faster.

…but thats not what the OP is talking about, and its not what I see either:

Its people putting up AI slop and not caring at all what the content was.

Just a quick check it compiled and the tests pass if youre lucky.

Too lazy even put a “dont use classes” in their cursor rules file.

Come on. The OP isnt saying dont use AI.

Theyre saying care, just a little bit about your craft ffs.

noodletheworld··on Do not download the app, use the website
I feel like many web developers want this to be true, but it is categorically false.

When you target a higher level abstraction, be it web, or flutter or whatever, you are explicitly choosing not to follow the platform native UX.

It’s more convenient to developers not to have to worry about that.

That’s it.

Web is easy. It’s free.

That doesn’t mean it’s better, or that it’s even possible for it to be as good as a native experience.

You can make a web app that is good; but it is the unavoidable and undeniable reality that web applications have a glass ceiling.

It is. Not. Possible. to write a web app that is as good as the equivalent native application can be. Certainly not a cross browser one.

There are reasons, you can blame Apple and safari or whatever you want, but that’s where it’s at, today.

> The reason I believe the web experience is inferior is because companies put more resources into apps at the expense of the web.

It’s not a falsifiable argument.

“That is not as good because I believe less effort was put into it”.

Ok.

I believe that for the equivalent effort you could create a better web app than a native app. I think you could measure that, and it would be pretty clear.

However, I believe many large native applications could not be implemented using the web platform. I think react native and the disaster that is is a reasonably solid proof that this is true.

They’re worse because web is worse, not because they didn’t bother to put effort in; because it wasn’t possible to do it using the web platform.

Native is always better if you out the effort in. It has capabilities that web doesn’t.

It is impossible for it not to be better.

noodletheworld··on US AI Action Plan
“I don’t know if they’re woke”

Yes you do. They’re not.

> I think people vastly overestimate their efficacy and efficiency

Of course, you can argue that people doctor the numbers (for example, failing to take into account the lifetime cost of nuclear power, or failing to note how hopelessly optimistic a pure solar power grid with no batteries might be) when they present said numbers… but the idea that any kind of power generation can be “woke” is beyond belief.

That isn’t an adjective that can be applied to physical processes.

Solar power is not woke. Gravity is not woke. Electricity is not woke. Don’t be daft.

noodletheworld··on Nobody knows how to build with AI yet
It doesn’t matter what downtime task it is.

You won’t have time to do it; it’s naive and ridiculous to expect that businesses will just let people goof off for 7 hours a day.

Regardless of the output they generate.

Anyone who doesn’t believe this has never had to manage budgets and staff.

It’s the “AI utopia” people making vague hand wavey motions about post-scarcity.

noodletheworld··on Nobody knows how to build with AI yet
There is a fundamentally unfortunate reality here which is quite problematic.

Namely, you don’t deserve to be paid for working 8 hours if you only worked for 30 minutes over an eight hour period.

I don’t care if you personally agree with that or not, the reality is that businesses believe it.

That means, sooner or later there will be a great rebalancing where people will be required to do significantly more work; probably the work of other developers who will be fired.

It’s fun for home projects; but the somewhat depressing reality is that there is no chance in hell this (sitting around for 7 hours a day reading reddit while Claude codes) will fly in corporate environments; instead, you’re looking at mass layoffs.

So. Enjoy it while you can folks.

In the future you’ll be spending that 8 hours struggling to juggle the context and review 20 different tasks, not playing with your kids.

noodletheworld··on The borrowchecker is what I like the least about Rust
> If you go out of bounds, Rust will panic and tell you exactly where it panicked

This arguement has a long history.

It is a widely used pattern in rust.

It is true that panics are memory safe, and there is nothing unsafe about having your own ref ids.

However, I believe thats its both fair and widely acknowledged that in general this approach is prone to bugs that cause panics for exactly this reason, and thats bad.

Just use Arc or Rc.

Or, an existing crate that implements a wrapper around it.

Its enormously unlikely that most applications need the performance of avoiding them, and very likely that if you are rolling your own, youll get caught up by edge cases.

This is a prime example of a rust antipattern.

You shouldnt be implementing it in your application code.

noodletheworld··on Writing Code Was Never the Bottleneck
> Why can't the same methods work post-llm?

>> If you don’t think about the implementation, or don’t understand the implementation, I would argue that you can earnestly try to test, but you won’t do a good job of it.

Previously the producers of the code were competent to test it independently.

This increasingly, to my personal observation, appears to no longer be the case.

They do test it, they just dont think about it deeply and so they do a shit job of testing it, and an incompetent job of writing tests for it.

Not by being lazy; smart diligent folk doing a bad job because they didn't actually understand what needed to be tested, and tested some irrelevant trivial happy path based on the requirements not the implementation.

Thats what LLMs give you.

Its not a process issue; its people earnestly thinking they've done a good job when they havent.

noodletheworld··on OBBB signed: Reinstates immediate expensing for U.S.-based R&D
Are you being serious or sarcastic? I cant tell.

Seriously, that seems unlikely.

Changes like this may have an impact on employment but it’s impossible to observe the results in a vacuum.

Given that most large companies are towing the “AI means less jobs required” line, it seems likely that this will, at best, modestly slow the rate at which companies divest themselves of software developers.

I cant see any reasonable reason, in a broader context, this would have a meaningful impact.

(Yeah yeah, AI means more jobs one day maybe, but right now that is categorically not true, and the future is always pure speculation, but in the near term, the impact of this seems like it probably wont be material to me; maybe a small reduction in the number of layoffs)

noodletheworld··on Writing Code Was Never the Bottleneck
How do you test edge cases?

You think about the implementation and how it can fail. If you don’t think about the implementation, or don’t understand the implementation, I would argue that you can earnestly try to test, but you won’t do a good job of it.

The issue of LLMs here is the proliferation of people not understanding the code they produce.

Having agents or LLMs review and understand and test code may be the future, but right now they’re quite bad at it, and that means that the parent comment is spot on; what I see right now is people producing AI content and pushing the burden of verification and understanding to other people.

noodletheworld··on I'm dialing back my LLM usage
> want to tell you how awesome their workflow is

Or, often, sell you something.

noodletheworld··on P-Hacking in Startups
“This kind of thing” being running AB tests at all.

There’s no reason to run AB / MVT tests at all if you’re not doing them properly.

noodletheworld··on P-Hacking in Startups
There are many resources that will explain this rigorously if you search for the term “p-hacking”.

The TLDR as I understand it is:

All data has patterns. If you look hard enough, you will find something.

How do you tell the difference between random variance and an actual pattern?

It’s simple and rigorously correct to only search the data for a single metric; other methods, eg. Bonferroni correction (divide p by k) exist, but are controversial (1).

Basically, are you a statistician? If not, sticking to the best practices in experimentation means your results are going to be meaningful.

If you see a pattern in another metric, run another experiment.

[1] - https://pmc.ncbi.nlm.nih.gov/articles/PMC1112991/

noodletheworld··on MCP Specification – version 2025-06-18 changes
No.

That (prompt injection) isn’t something you can fix until you come up with a way to split prompts.

That means new types models; there is no variant on MCP that can solve it with existing models.

noodletheworld··on LLMs are cheap
…and its perfectly legit to run that, write the numbers down and link to it.

But:

A) it makes absolutely no difference to the fact you have no idea what the big LLM providers are actually doing.

B) Just asserting some random thing and saying “anyone competent can verify this themselves” is a weak argument. Youre saying youve done the research, but failing to provide any evidence you actual have

If youve crunched the numbers then man up and post them.

If not, then stop at “I think…”

“This is based on my experience running production workloads…” is a nice way of saying “I dont have any data to backup what Im saying”.

If you did, you could just link to it.

…by not posting data you make your argument non-falisifyable.

It is just an oppinion.

noodletheworld··on LLMs are cheap
I don’t completely disagree, but “assertion one” [1]

[1] ~ you can obviously verify this yourself by doing it yourself and seeing how expensive it is.

…is an enormously weak argument.

You suppose. You guess. We guess.

Let’s be honest, you can just stop at:

> I don’t think so.

Fair. I don’t either; but that’s about all we can really get at the moment afaik.

noodletheworld··on Cloudlflare builds OAuth with Claude and publishes all the prompts
If you had written it by hand would the verification process been as time consuming?

i.e. overall including the time spent verifying that it was correct, do you consider it a net win?

noodletheworld··on Run a C# file directly using dotnet run app.cs
It is reasonably unlikely that bash scripts are easily replaceable by powershell scripts.

Theres a fair argument that complex scripts require a complex scripting language, but you have to have a good reason to pick powershell.

Typically, on non-windows, there is not one.

Its the same “tier” as bash; the thing you use because its there and then reach past for hard things. The same reason as bash.

Theres no realistic situation I would (and Ive written a lot of powershell) go, “gee, this bash script/makefile is too big and complex, lets rewrite it in powershell!”

noodletheworld··on MCP is the coming of Web 2.0 2.0
This is correct imo.

...and just like no one is prepared to pay 500 different vendors for micro-transactions, no one is prepared to pay 500 different websites for their MCP services.

Much much more likely is that a few big players (like google and AWS) will have paid-tier 'mega-MCP' servers, that offer 90% of what people need and fit in with existing payment/auth solutions (like your AWS account) and...

...everyone else with an MCP server will be left out in the cold.

That's what's going to happen.

These MCP servers will be cute for a very very short amount of time until people try to monetize them, and then they will consolidate very very quickly into a much smaller set of servers that people are already paying. Mostly cloud providers.

noodletheworld··on Diffusion models explained simply
Mmm… how is a model with a fixed size, let’s say, 512x512 (ie. 64x64 latent or whatever), able to output coherent images at a larger size, let’s say, 1024x1024?

Not in a “kind of like this” kind of way: PyTorch vector pipelines can’t take arbitrary sized inputs at runtime right?

If you input has shape [x, y, z] you cannot pass [2x, 2y, 2z] into it.

Not… “it works but not very well”; like, it cannot execute the pipeline if the input dimensions aren’t exactly what they were when training.

Right? Isn’t that how it works?

So, is the image chunked into fixed patches and fed through in parts? Or something else?

For example, (1) this toy implementation resizes the input image to match the expected input, and always emits an output of a specific fixed size.

Which is what you would expect; but also, points to tools like stable diffusion working in a way that is distinctly different to what the trivial explanation tend to say does?

[1] - https://github.com/uygarkurt/UNet-PyTorch/blob/main/inferenc...

← PreviousPage 6 of 8Next →