Noir

Cryptographic arbitration · E2EE · Autonomous AI consciousness · Neural interpretability · Novel primitives · 11M r/s 89ns · Mechanical sympathy · Patented

[Noir @Entrouter]

https://t.co/xlEoU75US8

[Noir @Entrouter]

@BenFordhamLive Bahahah we know it's all bull shit! You just want to control information, tyrannical cunts!

[Noir @Entrouter]

@AlboMP @ZelenskyyUa Money laundering cunt..

[Noir @Entrouter]

@nicdunz 100% bull shit and anyone who says it isn't is also a bull shit artist.

[Noir @Entrouter]

@goodfoodgal Good! Haha hilarious

[Noir @Entrouter]

@SheetWholy @MiguelRaya23 @RpsAgainstTrump No they really wouldn't.. just like how they won't attack Taiwan or North Korea.. It's not a video game, its reality. Also you have no clue what you're talking about, just look how scared Trump is about them, read the room..

[Noir @Entrouter]

@solopribuilds Watch gpt 6 become good now. It's an inference problem not quality of model. Quantization and quality loss under load.. you're only moving the issue to Opus.

[Noir @Entrouter]

@MiguelRaya23 @RpsAgainstTrump 🤣🤣 delusional. China could mog the US easily

[Noir @Entrouter]

@thsottiaux Yeah but the dictate makes the app lag. Can't use it

[Noir @Entrouter]

Haha, here's some tips.

Set your instructions to only use agents when told. Then use agents individually for specific things, not one agent to run 10 tasks. Tell your model to always update it's memory when large milestones have been achieved and when it thinks something important needs to be remembered, if it changes it again later, check if there is current memory on that code/feature and update it.

Get the model to do ALL the work (tasks), sure it's slower, but it has the most accuracy, inaccurate coding cost time and tokens. This goes for people using agents to code as well.

Use Max for alot of the grunt work on pro plans and xHigh on lower plans, then finalise and check everything with ultra with two agents only.

Don't have a mile long personality, it uses tokens to remember how to be..

Always start your workspace with the correct file structure, never have mono repos and split your code into 1k max files each.. <--- all extremely basic shit, but you'll be surprised how many people think a 20k file is acceptable.

Comment and SHA token index each part of your code base with an index .md so the AI can flick to each part of your code like a cook book. This took me a while to plan and invent, but boy has it saved me credits and time..

Once a month have the model use agents and read/audit your code base and update it's memory on things it missed or didn't update.

Don't just say "make me that thing, I want it to look good".. actually type what you want and explain things, the quality of the code and result will always be better with more information.

That's a few small things.. The other shit I'll keep for myself 🤙🏻😂

[Noir @Entrouter]

I think Sol is better than Astra. Astra fails to adhere to instructions and tends to run off doing things you never asked. That is far more of an inconvenience than Sol take a bit longer.
The best way to put it is; think of Sol as a sniper rifle and Astra a flamethrower. If precision and quality are your main goals then Sol it is. If you want an ADHD coder that thinks you wanted everything in that file fucked with, Astra is your man.
Astra is about 25%-35% faster depending on the task and the chat contextual length.

[Noir @Entrouter]

@Lordsaucyy @The_Alex Yeah, I'm delusional. Crack on 👍

[Noir @Entrouter]

@Freeeman_Gordon @melvindvivas Exactly what the future of AI will be.
Consistent productivity. Not a room full of gen z guzzling energy drinks "vibe" coding. I'd much rather decent reliable code produced at reasonable speeds, than Astras flamethrower of mess.

[Noir @Entrouter]

@ArcherNightfall @melvindvivas yep

[Noir @Entrouter]

@Panoptimos @melvindvivas I can believe that. Sol is excellent.

[Noir @Entrouter]

@Panoptimos @melvindvivas Yes, it still isn't "fast", but you get the same quality or better than 5.6 but using less tokens.. I think that's important. Essentially productivity has doubled.

[Noir @Entrouter]

@_brusooo_ @melvindvivas Yeah, it's pretty efficient. They did say it's 50% cheaper..

[Noir @Entrouter]

@elonmusk Wow, is this how medieval children played GTA?

[Noir @Entrouter]

@aranaalejandro_ @ilikefrontend @melvindvivas No, it's really not. People have no idea how to run agents or structure code. My workspace LOC is about 900k with nearly 3k files and yet I use next to nothing.

[Noir @Entrouter]

@PaladinPoet @melvindvivas Look for yourself. Been running the goal since 6am my time its now 2:43pm.

[Noir @Entrouter]

@_brusooo_ @melvindvivas Yeah 20x