nateb2022

Love the idea but I can't help but think of this as RCE as a feature haha

> For comparison, DeepSeek V4 Flash 0731, a 284B model, costs around $0.1/Mtok for output from the cheapest providers on OpenRouter. I am not sure how much of this difference comes from the efficiency of running models at scale, pricing strategy, or popularity.

Due in large part to DeepSeek's MoE architecture. Generating 1 token through DS4's 13B active params requires roughly half the FLOPs to generate a token through the dense Qwen 27B.

I would love to see some "pro" features hit the Pro or Pro Max. Something I've wanted for a while is the ability to have more than 2 active eSIMs; I think a dual modem would be really cool, maybe with some features targeting users in poor cell coverage/ spotty service areas.

Or feature Thunderbolt rather than normie USB-C so you can record 4/8K ProRes to an SSD, and power a multi monitor setup with your iPhone Pro. macOS would be sick. Or (waterproof) exposed PCIE lanes so 3rd parties can integrate an ecosystem of backpacks like how Pi Hats work.

> HN spread total disinformation and propaganda about it to us, blaming governments who actually banned it

Source?

cf. https://github.com/TellinaTool/nl2bash

paper: https://arxiv.org/abs/1802.08979

Interesting, it blocks 3 out of my 4 domains. One of the blocked ones is over a year old.

2 (1 blocked, 1 not) use Cloudflare for DNS. The other 2 use porkbun.

I'm surprised the OpenAI employee who pushed this didn't take the minute or two to format README.md to use GitHub-supported LaTeX (https://docs.github.com/en/get-started/writing-on-github/wor...)

edit: my comment was on the submission for https://github.com/openai/PrimeGaps186 but seems to have been moved to the main Astra submission

The article had been posted on the openai subreddit, which was live at time of posting (giving me a 404 now). Title copied verbatim from article.

US v. Wong Kim Ark (1898) has much bearing on this question and remains one of the best precedental interpretations of the 14th Amendment wrt this issue; It records the matter in which:

  The question presented by the record is whether a child born in the United States, of parents of Chinese descent, who, at the time of his birth, are subjects of the Emperor of China, but have a permanent domicil and residence in the United States, and are there carrying on business, and are not employed in any diplomatic or official capacity under the Emperor of China, becomes at the time of his birth a citizen of the United States, by virtue of the first clause of the Fourteenth Amendment of the Constitution, “All persons born or  naturalized in the United States, and subject to the jurisdiction thereof, are citizens of the United States and of the State wherein they reside.”

(https://www.govinfo.gov/content/pkg/USREPORTS-169/pdf/USREPO...)

And I quote:

  The interpretation of the Constitution of the United States is necessarily influenced by the fact that its provisions are framed in the language of the English common law, and are to be read in the light of its history.

and further:

  The fundamental principle of the common law with regard to English nationality was birth within the allegiance, also called “ligealty,” “obedience,” “faith” or “power,” of the King. The principle embraced all persons born within the King’s allegiance and subject to his protection. Such allegiance and protection were mutual — as expressed in the maxim, protectio trahit subjectionem, et subjectio protectionem — and were not restricted to natural-born subjects and naturalized subjects, or to those who had taken an oath of allegiance ; but were predicable of aliens in amity, so long as they were within the kingdom. Children, born in England, of such aliens, were therefore natural-born subjects. But the children, born within the realm, of foreign ambassadors, or the children of alien enemies, born during and within their hostile occupation of part of the King’s dominions, were not natural-born subjects, because not born within the allegiance, the obedience, or the power, or, as would be said at this day, within the jurisdiction of the King.

among other things, upon which it was concluded that

  The foregoing considerations and authorities irresistibly lead us to these conclusions: The Fourteenth Amendment affirms the ancient and fundamental rule of citizenship by  birth within the territory, in the allegiance and under the protection of the country, including all children here born of resident aliens, with the exceptions or qualifications (as old as the rule itself) of children of foreign sovereigns or their ministers, or born on foreign public ships, or of enemies within and during a hostile occupation of part of our territory, and with the single additional exception of children of members of the Indian tribes owing direct allegiance to their several tribes. The Amendment, in clear words and in manifest intent, includes the children born, within the territory of the United States, of all other persons, of whatever race or color, domiciled within the United States. Every citizen or subject of another country, while domiciled here, is within the allegiance and the protection, and consequently subject to the jurisdiction, of the United States. His allegiance to the United States is direct and immediate, and, although but local and temporary, continuing only so long as he remains within our territory, is yet, in the words of Lord Coke, in Calvin's Case, 7 Rep. Qa, “ strong enough to make a natural subject, for if he hath issue here, that issue is a natural-born subject; ” and his child, as said by Mr. Binney in his essay before quoted, “if born in the country, is as much a citizen as the naturalborn child of a citizen, and by operation of the same principle.”

and

  To hold that the Fourteenth Amendment of the Constitution excludes from citizenship the children, born in the United States, of citizens or subjects of other countries, would be to deny citizenship to thousands of persons of English, Scotch, Irish, German or other European parentage, who have always been considered and treated as citizens of the United States.

Regardless of whether birthright citizenship is appropriate contemporary policy, changing it would require an Amendment to the Constitution.

I would contend most corporations have a lot more managers than is necessary.

You can take a more involved approach to AI assisted development. Work at the line/function completion level, rather than vibing inside a harness. Zed offers a lot of this style of development out of the box, and you can use Copilot inside VSCode for a similar experience, or use a regular harness inside a terminal pane in VSCode but discipline yourself to small, targeted edits. As a human you have the strongest right to be the loop. Be the orchestrator.

Exactly this. The point of human management is not to do subordinates' jobs, rather to ensure that subordinates are performing their jobs to some standard of adequacy.

Claude went down earlier today

Yeah, I would be impressed by an economist talented enough to discern a defensible signal on the impact of AI on real job growth. I think the next best thing we have is to observe the tendencies of supply/demand; if people were actively being replaced by AI, that large demand would be being enabled by some pretty wealthy companies providing AI laborers, white collar desk job perhaps; however the companies providing such a service still involve a lot of human-in-the-loop type stuff, like those startups offering "AI coworkers" through a Slack or Teams interface.

Those companies though are not nearly as successful as OpenAI/Anthropic last I checked, which are supplying a lot of consumer AI/ AI being used by humans rather than AI doing what humans do, independently. And I'm pretty sure the majority of AI token burn is not replacing humans; it's building software under the direction of humans.

Correlation isn't causation; in the last few years we've experienced some significant economic headwinds like COVID and the current oil crisis. AI rollout is not happening in a vacuum.

I agree there, I think if we differentiate between first and second order AI companies (companies producing models vs applying those models to everyday tasks) I do think a lot of the hype comes from the 2nd order companies, since they necessarily must make a bold claim to be worthy of VC money. A lot of the 1st order companies are marketing their AI to consumers who are doing work, rather than looking to market their AI to business who are replacing those consumers. If anyone has the means to replace a worker, it's definitely OpenAI or Anthropic moreso than a 2nd order startup trying to replace the human doing the job.

And on a restrospective point, I think one of the easiest jobs for AI to do right now is software engineering. But on the whole, while there certainly have been layoffs, software engineering hiring has still been mostly a net positive[0].

And that's just for formal software engineering jobs; if we define a software engineer as a human creating software, then I think AI as a tool has created a LOT more (perhaps not employed as one, but say, a realtor creating their own sofware for personal use) software engineers than the amount that it's actually replaced. In a sense, because of AI, there are more software engineers now than ever before.

[0]: https://newsletter.pragmaticengineer.com/p/state-of-the-job-...

There's a lot of that type of slop on LinkedIn (with accompanying slop comments) but they're not taken anywhere as seriously as the more thoughtful blog posts like this

Also in technical circles, we're pretty aware of the marketing/self-promotional nuance of the "our AI accidentally hacked X company" type of press releases, which perhaps the general public may not take with as large a grain of salt

Most of the alarmist writing I see about AI displacing workers or suddenly and drastically changing the economy seem to be written by people who are pretty removed from technology or the actual businesses in which AI is allegedly replacing people.

Their theses on AI are not as much a criticism of AI as much as a desire to frame capitalism itself within their own system of ethics and advocate for some form of reform; which is to say I feel that were AI as we know it to be inexistent, they would still find themselves arguing against technology in general, or more precisely against efficiency, with a presupposed socialist-adjacent premise that capitalism only incentivizes greed.

Your 2x-4x cheaper claim is really hard to verify without transparent pricing for input/cache/output. Seems like you don't want to be easily compared to other providers, like on OpenRouter?

Edit: some back of the napkin math on GLM 5.2:

  $51.20 for 1B input, 4M output, 98% cache rate
  $303.20 for 1B input, 4M output, 83% cache rate
  $329.60 for 1B input, 10M output, 83% cached
  => $1.68/M input, $0/M cached, $4.40/M output

Which seems more expensive compared to $0.4186/$0.07774/$1.316 in/cache/out on OpenRouter: https://openrouter.ai/z-ai/glm-5.2

Tangentially this makes me wonder how large Opus really is. Perhaps Opus is a lot smaller than most of the 1T+ assumptions, just a lot more post-training/ finetuning on a 300-400B sized MoE model.

The $299 introductory price is not bad at all; for comparison, a Raspberry Pi 5 with 16GB RAM runs you ~$350 (Adafruit). And this Ventuno Q includes 40TOPS that the Pi doesn't come with by default, the Pi AI HAT+ 2 costs another ~$220 (Cyton.io).

So $299 vs $570, and the Arduino also has 4 more cores.

pyreqwest has been delightful so far: https://github.com/MarkusSintonen/pyreqwest

It's been a pleasure to use, has a httpx compatibility layer, and it's fast (https://github.com/MarkusSintonen/pyreqwest/blob/main/docs/b...)

I think the weighted average takes into account all providers (some DS4 flash providers are 'premium' providers and offering higher speeds for higher pricing) and these are tilting the scale

updated thanks

Slightly more expensive than the (post-price hike) DS4 flash pricing, but in the ballpark.

https://openrouter.ai/compare/deepseek/deepseek-v4-flash-073...

Curious why they're using 3.6-27B and not 3.8-27B which is competitive with Opus 4.6 (https://huggingface.co/Qwen/Qwen3.8-27B)

With an astonishing 160 commits on July 18th to this one repository (see https://github.com/taylannuhogluofficial-png), and no other activity in any other repository, I would be cautious of this seemingly benign offering.

Also I don't see much evidence/proof of care; e.g. there are hallucinations such as a reference to MITRE T1059.007 (https://github.com/taylannuhogluofficial-png/Tblue/blob/208b...) as "Client-Side SQL Injection (Web SQL)" when actually, it deals with JavaScript and not Web SQL: https://d3fend.mitre.org/offensive-technique/attack/T1059.00...

The author's AI applied a thin layer of varnish to what is more or less handwaving. I think more people need to take an undergraduate level statistics course; the conclusions made are completely unsupported by the random trials.

Weighing six rocks a hundred times doesn't tell you the average weight of a mountain.

> ds pro and flash are fast but not something id actually use for important things, unlike sol, fable and maybe 3.7 here and there

As someone who's used Gemini 3.7 Flash (Google sub mostly for the storage) and DS4 Flash a lot (~6B tokens), I'd actually place DS4 Flash (even pre-0713) above Gemini 3.7 Flash. Gemini has a tendency to leave some things unimplemented; perhaps it's agy which frankly leaves a bit to be desired as a harness.

Although I will praise DS4 Flash any day, it no longer makes sense for me after the price increase (GPT 5.6 Luna is a much better price point) and I have completely migrated my high volume workflows to Muse Spark 1.2 Contributor (which I find to perform better than DS4 Flash 0713, happily).

Well OpenRouter went from ~$5M in mid 2025 to ~$140M about a year later, so although 50x might be compared to current revenue, they probably factored in some growth models to approximate its value over time.