_flux · HackerTrans

11 घंटे पहलेdiscuss

The conversions seem to use (sometimes?) unsafe blocks.

परसोंdiscuss

Definitely mypy keeps my sanity in a relatively large Python project I'm working on :). But I don't believe Python actually assigns any meaning to them, and at times mypy and e.g. pyright may disagree on what the types actually are.

_flux

3 दिन पहलेdiscuss

There's a school of thought that consider the term "types" reflect to the properties that exist in programs even before they are run, as in they are a property of the programs themselves, not their state at runtime. This thinking—which is also what type theory talks about—does consider Python untyped: reading a Python program along with its specification, you are not able to assign types to each expression.

But what Python does have is tagging: when you create an object you tag it, and then whenever you operate on those values, you check the tag and maybe raise an exception or not. This is happening at runtime.

Strongly typed and weakly typed do not seem to have good definitions. A good one I've read is that "strong typing describes the typing you like".

It is great though if people go to the same extent as you to define what they are talking about, as this reduces the chances of misunderstandings. But it should not be taken as fact that the definitions you have chosen are the universally accepted ones.

_flux

3 दिन पहलेdiscuss

Could have used bigger font.

_flux

3 दिन पहलेdiscuss

You might not be able to express the ownership in the way that can be checked statically, so quite possibly this would then be downgraded to a runtime error (that could be handled with a panic)—but not undefined behavior.

_flux

3 दिन पहलेdiscuss

I wonder how many of the Linux the LPEs are related to drivers, which I understand there are more of..

_flux

3 दिन पहलेdiscuss

On one hand it's nice how it's clean and commented, but on the other hand some golfing could have made the encoded block a lot more reasonable to actually manually enter.

_flux

5 दिन पहलेdiscuss

It could use array microphone to detect that the sound originates from the driver's seat (in addition to using it for filtering out not-from-driver's-seat sounds).

_flux

5 दिन पहलेdiscuss

Paraphrasing: if I don't write the codebase [but someone else does], I don't write the tests, and other team members are modifying it, what even is considered a messy codebase?

I actually don't see a connection between the mechanism used to create the code and the code messiness. Things like code repetition, incorrect level of abstractions, tests testing only tests themselves, using too smart optimizations for things that don't matter, .. These all can happen in both person- and machine -authored code.

I would be surprised if a professional software developer has never seen at least some aspects of messy codebase in most any large project. The difference can be whether this messiness ever managed, or just piled on.

_flux

8 दिन पहलेdiscuss

Does Alibaba only have developers in the China?

_flux

8 दिन पहलेdiscuss

Wouldn't it then be reasonable to focus on those many features, instead of storage? I would enjoy using with S3, as expanding S3 storage is easier than expanding the storage of a virtual machine: usually it happens automatically.

Of course this topic has been discussed: https://github.com/immich-app/immich/discussions/1683

_flux

9 दिन पहलेdiscuss

You can't assign the copyright to Emacs either, yet it can be used to produce software.

_flux

12 दिन पहलेdiscuss

Works for me (TM). Maybe you lost CDN-lottery?

_flux

12 दिन पहलेdiscuss

Yes, but one only needs to write it once, and then everyone could use it. It could probably even be packaged as an official Debian package.

_flux

12 दिन पहलेdiscuss

Hmm, I perhaps didn't quite catch this. I thought having a generator would let you be less committed to it, as you wouldn't need to manually write all the init scripts you need.. ?

_flux

12 दिन पहलेdiscuss

My /usr/lib/systemd takes 6.4 megabytes, so I think I could live with that overhead. Or I suppose just delete them manually.

What would actually be useful would be a generic OpenRC wrapper that would ingest those service files and provide traditional start/stop interface for them.

_flux

12 दिन पहलेdiscuss

Devuan us a separate Debian with its own repositories (and presumably many packages patched to improve systemd-less life), while this is just replacing Systemd with OpenRC on a Debian system, while keeping Debian repos etc.

_flux

15 दिन पहलेdiscuss

So I let ChatGPT do the legwork for me, but it does seem the price difference between inference for GPT-5.5 and open-weight frontier models like DeepSeek V4 Pro and Kimi K2.6, which both are smaller models and thus cheaper to run inference on, is only 8x or so.

Sources https://openai.com/business/pricing/#api says for GPT-5.5:

  Input:$5.00 / 1M tokens Cached input:$0.50 / 1M tokens Output:$30.00 / 1M tokens

and for https://docs.fireworks.ai/serverless/pricing DeepSeek V4 Pro:

  Input: $1.74 / 1M tokens Cached input: $0.145 / 1M tokens Output: $3.48 /

Ratios are: 2.8, 3.4, 8.6

So as these numbers seem reasonably comparable to SOTA, and the SOTA vendors have additional overhead, then I think it is fair to deem that the alternative explanation offered here is not the explanation:

> Why do you think that subscriptions are subsidized and not that enterprise tokens are sold at 3000% margin?

As it does seem like the GPT-5.5 API tokens do not have significant margin based on the overhead-free companies selling inference for smaller models at prices of the same scale, I think we can believe that the subscriptions must be heavily subsidized.

It should be noted though that DeepSeek itself sells this even cheaper, but they may also be in it for the getting market share.

_flux

15 दिन पहलेdiscuss

So they do not train models, and in addition their models are expected to be smaller than SOTA models, although we cannot know for sure by how much.

So what's the price difference, 3000x?

_flux

15 दिन पहलेdiscuss

I think this comes from the idea that serving these tokens without paying for training is already expensive, e.g. https://news.ycombinator.com/item?id=46613887 self-hosted solution might give you only 10-100x more affordable solution at cost.

So, given the SOTA providers with even larger models also need to continously be using considerable resources for training their next models, to fund future data centers, and make profit, the token costs are more likely reflecting the real costs, rather than the subscription costs.

comments