[–] 1 point 3 days ago* (last edited 3 days ago)

"Guide" is doing a lot of heavy lifting there. There's an infinity of difference between asking an AI to make something and using an AI to do the scutt work, looking over the code it made, running IRL tests, asking "why X? Why not Y?", swearing at it like a junior dev, asking it to fix a,b,c then finally getting acceptable code.

And even with all that - when you set it up correctly - it's still faster than doing it yourself (in certain circumstances).

It feels more like doing project management / "software engineering" than clacking the keys. It's amazing how different LLMs approach the same task though; I find Codex to be much more by the book than Claude, for example. The Qwen's are like Labradors - eager to please, clumsy, crash into things a lot. GLM and Kimi are different again. Etc.

You can't let anything just "make me X, make no mistakes". By the time you type in a ticket (with GO/NO GO gates, acceptance criteria etc) you might be looking at 500-1000 lines of text (which is why a re-usable template and procedures are important). You need repeatable procedures, acceptance tests - the lot.

And so it goes.

  • source
  • parent
  • context
  • Yeah, KaiOS is dead.

    Here is what I have - there must be a local version https://opelmobile.com.au/product/touch-flip/

    It's Android 8.1. I installed a different launcher, small browser (SmartCookieWeb) and a few other essentials (banking app, QR reader etc). If you don't mind the flip factor (the T9 keyboard is a dream on this one...and with TT9 Whistle, I have the best of all worlds)

    EDIT: Found it https://sunbeamwireless.com/product/sunbeam-f1-orchid

    US version is more locked down but jailbreak guides avail https://furbreeze.github.io/2025/01/23/sunbeam-f1-root.html

  • source
  • parent
  • context
  • [–] 0 points 3 days ago* (last edited 3 days ago)

    You're talking loud, but say so little.

    Your post history is an argument. You made it - in public. I’m just citing it.

    PS: I believe in you too, champ.

    PPS: Still waiting on well...frankly anything not resembling a brain fart from you. You can do it! Maximum effort!

  • source
  • parent
  • context
  • [–] 1 point 4 days ago (2 children)

    Yeah, that was my feeling on it too. Cute, but too itty bitty. I already have an android flip phone if I want that form factor.

    It sounds like the Cubots would be ideal for you...IF you can find evidence of CFW unlocking your regional bands.

    Sorry dude. Chin up - they nuked 3G here entirely. Which means none of the good, old phones work at all. I have a Nokia N95 that's awesome but I can't do squat with it as a phone any more.

  • source
  • parent
  • context
  • [–] 1 point 4 days ago* (last edited 4 days ago)

    I got mine off Aliexpress, fwiw.

    https://www.aliexpress.com/w/wholesale-duoqin-f22-pro-phone.html

    The phone itself is awesome. The T9 keyboard is ass (the membrane they used is just not good). I got mine in 2022 iirc. I cycle between my Samsung A20 (yes, it still works well, 9 yrs in) and the local equivalent of the TCL flip (android 8.1 based).

    I actually just installed this on it and it's blazing fast STT now -

    https://github.com/maxmilneaus/tt9-whistle

    Edit to add: I have the F21 (the smaller one, candybar). The F22 is about the same height and an iphone, except halfish of the width.

  • source
  • parent
  • context
  • [–] 1 point 4 days ago* (last edited 4 days ago) (7 children)

    Hmm. I think the issue with all these phones is that the modems can support those bands just fine but they're generally tuned for Asian and European markets. Predominantly because those sorts of phones sell en mass there, maybe.

    For example, I have a duoquin F22 pro (think modern Nokia). I know that folks over on XDA have flashed CFW to region unlock them for US.

    You might be able to find what you want (the Cubots sound like a near match) quite cheaply but you'll have to check XDA for CFWs first. Quite a few of these phones ship with unlocked bootloaders and AOSP by default so they are pretty easy to tinker with.

    Else... How do you feel about the JellyStar?

    https://www.unihertz.com/products/jelly-star

  • source
  • parent
  • context
  •  
    • *for certain definition of self hosted.

    Developed by Chinese open weights model dev Z.ai, GLM 5.1 isn't exactly a small model. At roughly 754 billion total parameters (40 billion activated for each token generated), the model requires roughly 380 GB of GPU memory just to hold the weights and another 6-26 GB per user to maintain context depending on precision.

    To serve GLM-5.3 at scale, Revolut deployed 56 Nvidia B200 GPUs, equivalent in GPU count to seven eight-GPU DGX B200 systems. But as Abitbol noted, the greatest performance gains didn't come from adding GPUs, but rather tuning the model's hyperparameters along with switching to NVFP4 quantization, implementing disaggregated prefill and decode, and tuning context lengths.

     

    https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits

    Starting in December, Plus, Pro, and Business plans’ usage allowance will be shared across ChatGPT and Codex.

    OpenAI confirms that:

    • ChatGPT and Codex will share a token-based allowance.
    • Instant chat will not consume this allowance.
    • More complex tasks and advanced models will consume more tokens.
    • Users can check their remaining allowance under Settings → Usage.
    • Users can purchase extra credits if they exhaust their allowance.

    TL;DR - ChatGPT conversations that use advanced reasoning will compete with Codex for the same allowance. And it will be token based, rather then message / turn based. How many tokens? Fuck you, that's how many. Want a coder and chat with a non brain dead LLM? Pick one.

    So, if you're thinking of re-upping your OpenCode Go or OpenRouter access (or you're fortunate enough to have a rig that can run Qwen3.8 27B, Flash or the soon to be arriving Qwen 4's), now's the time to fuck off jippity.

    Personally, I'm going to ditch it for OR credits and selective use of GLM, Kimi etc for code review. I don't have the rig for a local GLM :(

    Ride was always going to end but fortunately, open weight models are a thing.

    Now if I could only run Qwen3.8-27b at not minutes-per-token...

    view more: next ›