HNHacker News
TopNewBestAskShowJobs

anon373839

5,123 karma · joined May 21, 2023

submissionscomments
anon373839··on Token-maxxing is dead, OpenRouter's token usage plateaus
> Or maybe Codex/Claude Code got better and more people are just using those?

Anthropic’s revenue growth has leveled off, too. Not the exponential hockey stick that was predicted.

anon373839··on DeepSeek search and fetch – $0.75 per 1000 successful requests
Update: There is no such guarantee. Cloudflare does label some providers as ZDR. Others are non-ZDR (such as Exa). Cloudflare itself may be retaining request and response data.
anon373839··on DeepSeek search and fetch – $0.75 per 1000 successful requests
CloudFlare is: https://blog.cloudflare.com/introducing-web-search-api

But check the fine print. They talk about the providers guaranteeing ZDR, but I missed the part where they themselves do also.

anon373839··on DeepSeek search and fetch – $0.75 per 1000 successful requests
Well, any evidence would be better than nothing. All I saw was a blog post with some conclusory statements, but I didn't dig into this. (It wasn't a rhetorical question.)
anon373839··on DeepSeek search and fetch – $0.75 per 1000 successful requests
> We have actual evidence of DeepSeek using stolen credit cards

Do we? Anthropic made allegations, but have they substantiated them?

anon373839··on Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware
It really depends on which models you are comparing. These days, parameter count and architecture really don't do a great job of predicting the model's capabilities.

The reigning champion for a fast, high-intelligence LLM right now is Qwen 3.8 Flash Next at 125B + 50B of embeddings (can be offloaded to SSD for ~free), 6B active parameters. It's somewhere in the Opus 4.6-4.8 range, noticeably stronger than 3.8 27B, despite having only 6B active. Compare Meta's Muse Spark Open, which has 816B total and 42B active, and does not even reach the performance of Flash Next.

anon373839··on Ultramarine 1 Flash
Can I ask what kind of tooling you use around the model and for the UI layer? I’ve been looking for a well designed knowledge work harness to give nontechnical users but I haven’t really found one that I like. (I personally just use coding harness for this stuff, but that’s a poor fit for non-devs.)
anon373839··on Ultramarine 1 Flash
You were joking about the toaster, but Anthropic literally just wrote a blog post claiming that dangerous open source models could persist themselves to smart ovens. You can’t make this shit up.
anon373839··on OpenCode 2.0
What in the hell is wrong with this page? Whole blocks of content disappear and reappear on my phone as I scroll. Eventually the page just starts force reloading. It’s a blog post! What could possibly require … whatever this is?
anon373839··on Strengthening our defenses against distillation attacks
The horror! Replicating model capabilities without authorization.

Seriously though, these clowns at Anthropic are not just hypocritical. They are clearly feeling threatened by open source AI. I almost couldn’t believe the new blog post today where they argue that dangerous open source models might soon be running on (checks notes) smart ovens.

anon373839··on Gemini 4 Argon
Qwen Flash Next has great writing! Unless it comes into any contact whatsoever with Claudeslop - then it immediately contracts the disease. It’s kind of irritating when you have it look at library code that’s been touched by Claude, because then the rest of your session is infected with it.
anon373839··on GLM-Legal
This model would be self hostable quite easily (~$14k hardware investment) if it’s released as open weights, since the base model is GLM 5.3-Flash. If not, it’s completely DOA for legal use.
anon373839··on Research Firm Calls Anthropic's IPO 'Unprecedented Test of Investor Gullibity'
I don't see how Anthropic can become profitable enough to justify its investments unless they can either stop training models ("Pace the Frontier!!1"), or magically eliminate open source competition so that their volume x margin on inference is high enough to dwarf the training costs.

For the IPO, I assume they're going to try to get everyone to focus on inference - which is profitable in isolation. But I don't know how they'll distract the market from the other half of the equation.

anon373839··on Show HN: BetterWispr – Free, open-source dictation for Mac
Qwen 3.8 Flash Next also works very well with this, and frankly even the much smaller 3.6 35B-A3B did, too.
anon373839··on How Protein Took over the World
Now look up the amino acid profile of protein in white bread, though.
anon373839··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
> I'm absolutely bewildered that this is the top comment. 'Pacing the frontier,' if eventually enforced, will only affect US companies

I can address your bewilderment. What I was trying to say is that they don't actually want to pace the frontier at all. They want to gum up the market with regulations they design, that would ultimately force US companies to rent AI from them. They don't need to stop Chinese AI development and cannot do that. But they can, to quote OpenAI's Head of Strategic Futures, "create enough regulatory risk that every regulated enterprise backs off [of using Chinese models]".

https://x.com/deanwball/status/2078133895766114412

anon373839··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
Yes, I was being facetious. Chinese labs are doing the most interesting research and publishing it. But Anthropic beats the "distillation attack" drum every time doubts surface about the depth of their technical moat. Of course, when they need a scary bogeyman, then the story shifts to how China is recklessly racing ahead building dangerously powerful models. That's the thing with Anthropic: they speak out of all 13 sides of their mouths.
anon373839··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
Have you looked at this backwards, though? Meaning: have you considered what it would look like if the motivations are actually just plain old money/power, but a set of stated beliefs needed to be constructed to justify what's being sought? I think these absurd beliefs make more sense that way.
anon373839··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
> Sure, but does slowing down the development of new models count as "adopting safety standards"?

Of course! What slows down the development is the adoption of specific safety conditions the companies draw up. They can just do that, and it will hold up in court.

> in their view, letting other companies win would increase the probability of human extinction

Yes, I've heard: "We must be in charge even if we end up killing everyone in the process." I personally think that proposition is invalid, but we're all entitled to our opinions.

anon373839··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
> Antitrust law generally prohibits competing companies from agreeing to restrict innovation.

The antitrust claims are a complete smokescreen. Industries can and do adopt safety standards without government intervention.

> These companies are in a Nash equilibrium where they can't unilaterally slow down without essentially destroying their company.

So what? Anthropic believes their work has a 10% chance of killing all humans. I think risking the destruction of Anthropic's business should be worth avoiding that, if that's what they believe. And with one half of the frontier duopoly gone, the other half would have no incentive to race forward. And I know there's China, but they just get all their capabilities from distilling Claude, right? So, problem solved there, too.

anon373839··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
I think the industry is freaking out about open weights models in general, if not specifically DeepSeek. That is why we're now on the ~4th call for pacing the frontier from the very people who, if they wanted to pace the frontier, would simply do it rather than asking Washington to get involved.

And it's why people like Hillary Clinton have been trotted out to talk about the dangers of open weights models -- I mean, does she even know what that phrase means? (I know HRC is a controversial figure and I'm not bringing her up for that purpose; I just note that she and other prominent retired politicians are now doing the circuit on Anthropic's behalf.)

anon373839··on Anthropic reported diary entry to police, woman faces felony charge
Same here. I was planning to get a second, just got this one ten seconds before they became unobtanium. Here’s hoping Qwen 4 Flash is the same size!
anon373839··on Web Search API
> All three support Zero Data Retention for requests made through Cloudflare

But does CloudFlare itself commit to zero data retention? If not, this isn’t too meaningful.

anon373839··on Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s
Those prefill numbers don’t seem right for Qwen Flash Next. I get 3,000 tok/sec on a DGX Spark and that’s got a lot less raw compute than the 6000 Pro.
anon373839··on Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s
It is on paper, but crazy enough, both models at NVFP4 run similar speeds for decode! The reason is that much more sophisticated speculative drafting is available for 27B. I’m hoping this will come to Flash Next, but I know MoEs pose challenges with that.
anon373839··on Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s
Yep, can confirm that is NOT normal. Are you using Nvidia’s NVFP4 quant? There are other NVFP4s floating around but they are not as good. The quality of the calibration data really matters.

Qwen Flash Next is just excellent, all the way to the very end of the native 262k context. (I haven’t tried YaRN scaling to 1M, so I don’t know about that.)

anon373839··on As A.I. makes law firms more efficient, clients ask: 'Where's my discount?'
> Email and phone calls are considered confidential, even though it is possible for vendors to inspect the communication.

It's a weak analogy. Ordinary comms infrastructure providers actually have pretty robust policies, technical, and contractual measures in place that restrict employee access to customer communications. In contrast, in the wild west of generative AI, companies actively monitor session data for the content itself, in order to exploit it for their own business purposes. There is zero expectation of privacy.

So I don't share your expectation that precedent will uphold the use of consumer-tier services (in their current form) for handling privileged material.

anon373839··on As A.I. makes law firms more efficient, clients ask: 'Where's my discount?'
This is a completely straightforward application of existing law on privilege. To maintain privilege, among other things, communications must be confidential.

Chatting with Claude breaks confidentiality: chats with Claude are subject to arbitrary inspection by Anthropic employees, not to mention the issue of model training.

You can use self-hosted LLMs without breaking privilege. And funny enough, law firms like Latham & Watkins are now buying Nvidia GPU clusters for this purpose.

anon373839··on "As a Language Model": Chat Template Switches LLM Self-Referential Voice
God almighty, yes. Make it stop.
anon373839··on California is chasing wealth that has feet
> the argument that billionaires will leave if taxed at a higher rate isn't compelling

I agree with this. California’s climate and culture will keep many a billionaire within tax nexus reach of the state. Of course, this isn’t a strategy every locale can pursue, but I don’t see a reason for California not to exploit its advantages.

← PreviousPage 2 of 34Next →