Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Its on OR - but currently not available on their anthropic endpoint. OR if you read this, pls enable it there! I am using kimi-2.6 with Claude Code, works well, but Deepseek V4 gives an error:

`https://openrouter.ai/api/messages with model=deepseek/deepseek-v4-pro, OR returns an error because their Anthropic-compat translator doesn't cover V4 yet. The Claude CLI dutifully surfaces that error as "model...does not exist"


Whatever the capabilities, there’s always a little hype, or at least the risk won’t be as great as thought:

> Due to our concerns about malicious applications of the technology, we are not releasing the trained model.

That was for GPT-2 https://openai.com/index/better-language-models/


That is such an amazing idea. How does the pool of stories work? Is it the first in, first out principle, or does it pick a random one? What if a toy user listens through all the stories?

Search was not novel, but PageRank was novel.

Getting 'Api Error' here :( Every other model is working fine.

I think the impact of specific workspace arrangements will be different for each person. And no e of these adjustments are going to unlock a torrent of productivity.

From my own experience: * Getting an ear-covering gamer headset was most effective at blocking external sounds, and helping with focus * Replacing my cushioned work chair with a mesh bottom-and-back chair reduced sweating * Having an e-ink frame with certain reminders/affirmations helps me get through tough stretches at work


I was curious, and some [intrepid soul](https://wavespeed.ai/blog/posts/deepseek-v4-gpu-vram-require...) did an analysis. Assuming you do everything perfectly and take full advantage of the model's MoE sparsity, it would take:

- To run at full precision: "16–24 H100s", giving us ~$400-600k upfront, or $8-12/h from [us-east-1](https://intuitionlabs.ai/articles/h100-rental-prices-cloud-c...).

- To run with "heavy quantization" (16 bits -> 8): "8xH100", giving us $200K upfront and $4/h.

- To run truly "locally"--i.e. in a house instead of a data center--you'd need four 4090s, one of the most powerful consumer GPUs available. Even that would clock in around $15k for the cards alone and ~$0.22/h for the electricity (in the US).

Truly an insane industry. This is a good reminder of why datacenter capex from since 2023 has eclipsed the Manhattan Project, the Apollo program, and the US interstate system combined...


Now, I think the author would consider it "solutionism", but the other day I spent a bunch of time browsing Reticulum's NomadNet sites (using the Columba mobile app).

And while aesthetically it was more early 90s than 1999, it filled me not only with nostalgia, but also with some optimism for the future of the Internet. Something I haven't felt in a while...


I use glm and I like it, not they also increased the price to 18 usd /month.

I think Kimi and qwen are similar?


> For those unfamiliar, Slayer is (imo the best) one of the top $$$ machines and pairing it with a budget grinder is a classic sign the owner doesn't know a thing about coffee.

The exception to that rule is Espresso Vivace in Seattle, with (at Capitol Hill location) a couple 3-group La Marzoccos at the bar and a collection of modded Niche Zeros on grinding duty. Nobody can accuse David Schomer of "not knowing a thing about coffee".


Although in this case Meta bought companies that were already established and successful.

Google bought Android before it had released products.

Google Maps was purchased, but was Where 2 actually a successful product prior to that?


Guess who is expected to pay off these massive loans?

The users of AI --- who will be hit with equally massive bills.


What type of system is needed to self host this? How much would it cost?

> derivative markets will exist up to the divine wrath of god

We already bet on the weather.


Funny how Gemini is theoretically the best -- but in practice all the bugs in the interface mean I don't want to use it anymore. The worst is it forgets context (and lies about it), but it's very unreliable at reading pdfs (and lies about it). There's also no branch, so once the context is lost/polluted, you have to start projects over and build up the context from scratch again.

It's true in a case where you are doing the described thing you will need to come up with your own module system and ways to not step on other people's stuff but it isn't actually difficult. Although I have noticed some stories recently were quite big companies evidently didn't put in the work to keep from messing up other people's stuff.

Of course one drawback with that is you are also depending on developers and content managers at sites following your documentation on how to use your products, which is a different problem.


Minor correction from the article title: It can't stop ordering candles, not candies.

Although candies would've been more amusing


That's super interesting, isn't Deepseek in China banned from using Anthropic models? Yet here they're comparing it in terms of internal employee testing.

Aren’t they sort of unique in that they… have a retail division, as a real ongoing thing (I’m sure MS tried an MS store but I’ve never seen one).

Well, unique other than Amazon I guess.


Tibet is part of China for hundreds of years (1720–1912) with short period exemption (1912–1951), people who think China just suddenly invaded Tibet in 1951 out of blue are delusional or should learn history

btw. just because you hear loud minority (?) of Tibetan people unhappy with China's rule doesn't mean there is not big part of them who have no problem with benefitting from being part of China rather than let's say India/Nepal


Did you look at DBOS [0]?

It does everything you list here, and more, all locally. It's also MIT open source.

[0] https://github.com/dbos-inc/dbos-transact-py


if you want to understand why labs open source their models: http://try.works/why-chinese-ai-labs-went-open-and-will-rema...

At this point, I feel like there should be a HN post whenever there ISN'T an issue with some GitHub service. Otherwise, it's business as usual...

So, dual RTX PRO 6000

Meh, property tax rates are so high in Texas because there is no state income tax. And I disagree that prices aren't much lower in other towns. Whenever I drive to Houston I always comment when I stop for food (which usually ends up being at some "fast casual" type place) that prices are much more reasonable than in Austin, like ~20-25% less.

Vibes > Benchmarks. And it's all so task-specific. Gemini 3 has scored very well in benchmarks for very long but is poor at agentic usecases. A lot of people prefering Opus 4.6 to 4.7 for coding despite benchmarks, much more than I've seen before (4.5->4.6, 4->4.5).

Doesn't mean Deepseek v4 isn't great, just benchmarks alone aren't enough to tell.


Came here to reply with ImHex.

It really is by far the best hex editor I ever used, and sooo good for reversing arbitrary binary blobs where you learn incrementally more about its structure while reversing it. The imhex patterns repo [1] also contains so many formats, it makes binwalk almost useless in comparison.

[1] https://github.com/WerWolv/ImHex-Patterns


Yes, it also takes into account rising quality. For an example, in 2010 I rented a rat hole apartment for $x from a fisherman who had inherited the building. He never did maintenance (he was out to sea most of the year) and he never raised rent.

A large company bought the building after I moved out. Ten years later, the same apartment with a fresh coat of paint and new countertops was back on the market for a rent of about three times $x.

The CPI can say that apartment, since it was refurbished, increased in quality and so it wasn't really a price increase of the same good from $x to $3x. This offers a "degree of freedom" to adjust the CPI itself (since quality is inherently subjective), and may be a big part of why CPI does not reflect the lived experience.

I didn't care one bit about paint or countertops when I rented that apartment and I assume broke young adults today don't either. At the time I wanted the cheapest place to live in the area and this was it. It still is one of the cheapest places, but you need three times as much money to rent it.


Why people publish AI written articles? If I would like to read AI I can just prompt it myself, and when I read something on someone blog I expect that I will read thoughts of this particular human being...

Well, a lot of people got killed this way, too.

Consider applying for YC's Summer 2026 batch! Applications are open till May 4

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

Search: