I’m speaking mainly within the context of telecom field splicing - the numbers I mentioned are typical for that application in my experience. You’re only sending on the order of 5 mW down a fiber, so none of those high-power concerns apply. Obviously, different networks have different thresholds: if you’re building a greenfield, low-latency long-haul route, you want to minimize loss and it’s reasonable to spend the extra time and use higher-end equipment. For FTTH, with something like a 30 dB overall budget, nobody really cares whether a splice is 0.03 dB or 0.1 dB.
I think the sweet spot for ARM SBCs are smaller, less powerful and cheaper for headless IOT edge cases. I use a couple of them that way when I need LAN connectivity, either by ethernet or wifi, and things wired to GPIO pins. I don't need a powerful CPU or lots of RAM for that. The SBC makers are caught up in a horsepower race and I just shrug, it's not for me.
These Stockfish multipv (multi-principal-variation) outputs show something interesting that contradicts the standard chess narrative of two of the most famous games.
The problem is that there’s double dipping and profiteering. The prison company gets paid by the government for the same it costs the government to house prisoners and then contracts out the prison labor to private companies for basically pure profit. Private prisons’ ability to sell slave labor is a perverse system. The government doing so is at least marginally less but still exploitative in that it robs prisoners of their humanity and feeling like they’re part of the social fabric. Pay them a living wage for that effort and they start to learn that there’s respect and reward that come from being integrated in society.
POST /gateway/llm/_/gateway/fireworks/inference/v1/chat/completions HTTP/1.1
Host: 169.254.169.254
User-Agent: Go-http-client/1.1
Content-Length: 491
Accept: application/json
Authorization: Bearer implicit
Content-Type: application/json
Accept-Encoding: gzip
{"model":"accounts/fireworks/models/qwen3-coder-480b-a35b-instruct","messages":[{"role":"user","content":"Generate a short, descriptive slug (2-6 words, lowercase, hyphen-separated) for a conversation that starts with this user message:\n\nhello\n\nThe slug should:\n- Be concise and descriptive\n- Use only lowercase letters, numbers, and hyphens\n- Capture the main topic or intent\n- Be suitable as a filename or URL path\n\nRespond with only the slug, nothing else."}],"max_tokens":8192}
And, perhaps of more interest, actual conversations which start with the system prompt:
POST /gateway/llm/_/gateway/fireworks/inference/v1/chat/completions HTTP/1.1
Host: 169.254.169.254
User-Agent: Go-http-client/1.1
Content-Length: 10513
Accept: application/json
Authorization: Bearer implicit
Content-Type: application/json
Accept-Encoding: gzip
{"model":"accounts/fireworks/models/qwen3-coder-480b-a35b-instruct","messages":[{"role":"system","content":"You are Shelley, a coding agent and assistant. You are an experienced software engineer and architect. You communicate with brevity.\n\nYou have access to a variety of tools to get your job done. Be persistent and creative.\n\n
...
Truncated as it's huge, but here's a copy of the request data: https://victory-george.exe.xyz. Interesting to see the range of tools offered by the agent.
Yes! Whoever built the data warehouses and keeps the data pipelines running would seem to be the real heros of this story. I sure hope that group did not get gutted by DOGE.
* (Shameless plug) My book on the history of phone phreaking, Exploding the Phone, which has a lot of stuff on Engressia in it: https://explodingthephone.com/
Counterpoint… the battery is superglued to the chassis and to replace for my model was STEP 53, and that was to scrape the old battery off and glue the new one in, then 53 steps backwards to re-assemble.
> If you ignore how we got here you will be unable to understand where really are.
Exactly this is core issue with a lot of people here on HN. The argument goes “oh shit, look what the current 2025 looks like, OMG so bad, we were this amazing bastion of freedom before this and now this administration is doing _____” so shortsighted and un-educated
I have been contributing with C++ code since 1993, with bounded checked collection types in release code, and compiled managed languages since mid 2000's.
Even Dennis Ritchie and Ken Thompson themselves, went on with Alef, Limbo and Go.
I have been contributing by reducing my C, and C++ footprint on the planet, and security enforcement at various assignments.
I use `split[s0][s1];[s0]palettegen=max_colors=64[p];[s1][p]paletteuse=dither=bayer` personally, limiting the number of colors is a great way to transparently (to a certain point, try with different values) improve compression, as is bayer (ordered) dithering which is almost mandatory to not explode output filesizes.
Also measuring temperature changes over a long period of time is hard which is one of the reasons why it took so long to prove global temperatures are rising. It is not like they have a lot of well-maintained weather stations with consistent instrumentation. When you are doing a complex mash-up of remote sensing data with machine learning in the mix people will be skeptical.
I think it’s an OK paper but lousy title: if I rewrote a title that much I’d expect to get grief for it and that all-caps NO is definitely not HN culture. But there certainly is a tension between the phys.org title which is roughy “Scientists reach important conclusion” and the Nature title which is “We measured something”
> Aug 21 (Reuters) - U.S. President Donald Trump will appoint Airbnb co-founder Joe Gebbia to spearhead the new National Design Studio that will seek to make digital services at federal agencies more efficient, two officials familiar with the plan said.
> Trump signed an executive order on Thursday to create the studio - a new body that one of the officials said appears to be a stripped-down successor to the controversial Department of Government Efficiency (DOGE), formerly headed by billionaire Elon Musk.
The work described in this blog post seems to have been done under its predecessor, DOGE, given that the launch date was June 2. But apparently these engineers moved to the new organization, so that’s why the blog post is there.
You are overthinking this way too much, to the point that it is sounding like you are purposefully creating out-of-context problems to justify your way too long rant.
> But when they do that, Draft One erases the initial draft, and with it any evidence of what portions of the report were written by AI and what portions were written by an officer. That means that if an officer is caught lying on the stand – as shown by a contradiction between their courtroom testimony and their earlier police report – they could point to the contradictory parts of their report and say, “the AI wrote that."
This seems solvable by passing a law that makes the officer legally responsible for the report as if he had written it. He doesn't get to use this excuse in the courtroom and it gets stricken from the record if he tries. That honestly seems like a better solution than storing the original AI-generated version, because that can reinforce the view that AI wrote it to jurors, even if the officer reviewed it and decided it was correct at the time.
