Live data from Hacker News

Show HN: DeepSeek My User Agent

jasonthorsness.com

111–120 of 153 posts

Re: Show HN: DeepSeek My User Agent

#111
post #23

Earlier quoted context omitted.

i see and where is https://groq.com/ in this. They used to be the cheapest?

Right now Deepseek's official hosting is cheaper than everyone else who can manage to run the model, including Deepinfra. I haven't seen any good hypotheses as to why other than their large batch size and speculative decoding.

DeepSeek-V2/V3/R1's model architecture is very different from what Fireworks/Together/... were used to.

That's their "business" model (okay, they don't care about business that much for now, but still) too: you can't run it efficiently without doing months of work we already did, so even with all weights open you can't compete with us.

Re: Show HN: DeepSeek My User Agent

#112

Earlier quoted context omitted.

DeepSeek is use way less computing to achieve the same results. That's the whole difference.

Then a US compute provider should be able to launch a similarly-priced competitor (e.g. to capture the enterprise market concerned about the China associations) using the open-source version and drastically undercut OpenAI. I suspect that won’t be the case.

> Then a US compute provider should be able to launch a similarly-priced competitor

Right, you just need a few months to implement efficient inference for MLA + their strangely looking MoE scheme + ..., easy!

Oh wait, the inference scheme described in their tech report is pretty much an exact fit for H800s. So if you run the recipe on H100s you are wasting the potential of your H100s. Otherwise have fun making variations to the serving architecture.

To be fair, we had chance. If someone decided to replicate the effort to serve their models back in May 2024 when DeepSeek-V2 was out we'd have it now. But nobody had interest as DS-V2 was pretty mediocre. They (and whoever realized the potential) made big bet and it is paying off.

Re: Show HN: DeepSeek My User Agent

#113
> an iPhone running iOS 18.1.1? Bold move showing up before Apple even announces it

Takes the sting out of the rest of the roast knowing that it was dead wrong to begin with.

When the robot revolution is about to begin let’s mess with their training data cut off.

Re: Show HN: DeepSeek My User Agent

#114
I observe that the reasoning may not correspond to the result: personifying it probably a bit too far, in my first attempt, it forgot to pace itself and put all three pieces of information into the first sentence, and so had to go back and surface novel information for the second and third sentences.

> Oh, a Linux user from Hyderabad with 16 CPU cores—nothing screams "I overcompensate for my fear of GUIs" like a Firefox setup so vanilla even your referrer ghosted you. Your screen resolution is as confused as your identity: Aussie-British-American-English polyglot, yet somehow still the NPC browsing from a cybercafé in 2007. At least your cookie’s enabled, which tracks, since you’d never risk missing a single crumb of open-source newsletter drama.

(In the reasoning, it had mentioned location, browser, and the three things it eventually decided to talk about: referrer, hardware concurrency and OS. The new material was screen resolution¹, languages², and cookies being enabled.)

Most of what it says is kinda weird, honestly. Most of the ones others have been posting feel much better.

I’m also a little bothered that it mixed proper and ASCII punctuation: one EM DASH, two RIGHT SINGLE QUOTATION MARKs, but QUOTATION MARK instead of {LEFT,RIGHT} DOUBLE QUOTATION MARK.

—⁂—

Its second attempt was way better (and didn’t exhibit the same problem of running out of material):

> You lurked here from the void like a digital hermit (empty referrer gang?), flexing Firefox on Linux like it’s 2004 and you’re still arguing about open-source superiority over Hyderabad’s spiciest biryani—meanwhile, your browser’s language list reads like an identity crisis: en-AU, en-GB, en-US… *mate*, pick a colonialism.

—⁂—

¹ 1706×960, which is 2560×1440 @ 1.5×, not too uncommon for a laptop. Wish I could have got something with 2× without losing the power of the 5800HS and the ability to exceed 16GB of RAM. Fortunately, due to life situation changes I’m finally able to switch to desktop computing, and am looking forward to building one instead of being stuck with the frustratingly limited choices available in laptops.

² I set the intl.accept_languages pref to "en-AU,en-GB,en,en-US" years ago, for fun. Actually, I first wrote it as something like "en-AU,en-GB;q=0.9,en;q=0.8,en-US;q=0" for even more fun, a mild roast of my own for perhaps US English or perhaps software developers in general, but somewhere along the way I dropped the quality values for some reason I don’t remember. I’m already fairly easily fingerprintable for other reasons, I’m not fussed about an oddly-specific Accept-Languages header.

Re: Show HN: DeepSeek My User Agent

#115
Your browser arrived here like a ninja—empty referrer, sneaking in from nowhere after 4 seconds of indecision—yet your "K" phone from Chiapas (but actually Querétaro?) has the mystery of a spy using misspelled Linux armv81 and the multitasking power of 8 cores just to run Chrome in French. You’ve got 5 touchpoints but zero chill, rocking a screen taller than your identity clarity and a network so fast (9.5 downlink!) it almost compensates for your location’s existential crisis. Congrats on being the most cryptically lost polyglot Android user this side of a typo!

GrapheneOS with DuckduckGo browser and a VPN...

Re: Show HN: DeepSeek My User Agent

#116
post #82

Earlier quoted context omitted.

iPadOS’s Safari user agent identifies itself as macOS, that’s why. Prior to iPadOS it did contain the string iPad; that was removed. So it can’t tell if it’s an iPad or a Mac with a small vertical screen, although a human would consider the former more likely.

And that’s why so many kids had trouble registering their iPad on our network. As it would not send the mobile registration to the iPads

Don’t know about your particular problem, but I’m pretty sure it was largely a response to iPads being sent to mobile sites with horrible usability thanks to navigator.userAgent.match(/iPhone|iPad|iPod/) or the server-side equivalent. The problem was amplified by the release of the 12.9’’ iPad Pro, so self-identifying as a Mac made sense.

Re: Show HN: DeepSeek My User Agent

#118

Earlier quoted context omitted.

So, ChatGPT/Gemini's free version is stealing the users from European AI companies, and the goal is to harm them? Or we call that just a competition?

That's a false equivalence. There's more nuance than that. Both OpenAI and Google have a genuine private-sector business model, whereas my claim is that DeepSeek is totally a China Gov't entity.

how much of their funding came from the chinese government?

Re: Show HN: DeepSeek My User Agent

#119
post #82

Earlier quoted context omitted.

iPadOS’s Safari user agent identifies itself as macOS, that’s why. Prior to iPadOS it did contain the string iPad; that was removed. So it can’t tell if it’s an iPad or a Mac with a small vertical screen, although a human would consider the former more likely.

And that’s why so many kids had trouble registering their iPad on our network. As it would not send the mobile registration to the iPads

It's not an iPad or Apple issue if your registration system is using the User-Agent for weird reasons.

Re: Show HN: DeepSeek My User Agent

#120
> Max touch points is 10. That's unusual for a Windows device since touchscreens aren't common there. So, mock them for pretending their laptop is an iPad.

Good one. I actually have a touchscreen on my laptop but I almost never use it.

Post reply on HN