nick@BSLABS

[nick@BSLABS @F_AI_Mouse]

@ChatGPT we say "chi-bi-dee"

[nick@BSLABS @F_AI_Mouse]

I've got a new desktop landing in a few days and I'm excited to try Omarchy for the first time. I've been a big Linux fan for decades, but it's never stuck as a desktop for me. It's fun to feel the excitement in the air...

[Bjarne Øverli @iamdothash]

Do you remember when you first booted DOS and installed a game, e.g., Monkey Island, Lemmings, Worms, ... Larry?

Remember that feeling?

That is the feeling I have working on Omarchy.

[nick@BSLABS @F_AI_Mouse]

@ivanfioravanti Pairing local models will take us to interesting places. I have big plans for qwen 3.8 27b + Gemma 4 Diffusion... I'm not sure many have figured out how to make G4D sing - there's some alpha hiding in there

[nick@BSLABS @F_AI_Mouse]

@__tinygrad__ @scaling01 I'm so glad you guys finally got traction with AMD

[nick@BSLABS @F_AI_Mouse]

fond memories...

[exQUIZitely 🕹️ @exQUIZitely]

This is how a great weekend in the late 90s and early 2000s started.

Dude on the left is best prepared. Dude in the middle never skipped leg day. Dude on the right is playing with fire.

Yes, I know you can play everything online now. You don’t need to haul your gear halfway across the city.

But packing up your 17-inch monitor and your equally heavy tower to head over to your friend’s house and hang out all weekend… that feeling is impossible to replicate.

You had to be there. And if you were, I know you miss it.

[nick@BSLABS @F_AI_Mouse]

@no_stp_on_snek maybe I'm being overly simplistic, but couldn't you run an outer loop with abridged/compacted reasoning associated with each turn?

[nick@BSLABS @F_AI_Mouse]

Interestingly, I've found that the models really must be pushed hard to actually explore this way. Simply spelling it out in a skill or agent guidance just doesn't cut it. You have to force them to do it.

[nick@BSLABS @F_AI_Mouse]

I've spent the last few weeks refining a whole-repo caching system. If you are familiar with repoprompt, it's kind of like having the entirety of the repo contained within domain scoped repoprompt discovery contexts, with a spine that provides cohesion.

[nick@BSLABS @F_AI_Mouse]

It works beautifully when you interrogate the set of caches over multiple turns, tension between caches surface as you question them against each other, and as you chase the seams you discover issues that need resolution.

[nick@BSLABS @F_AI_Mouse]

Businesses in highly regulated domains will flock to SpaceXAI and grok4.x when everyone figures out you can get opus level models for 1/3 the cost with ZDR for free. OpenAI and Anthropic don't even post prices for ZDR, it's a 'call for pricing' situation with them.

[kache @yacineMTB]

I live in the future man

[nick@BSLABS @F_AI_Mouse]

@Code_X_Captain @ivanfioravanti I've been using codex app for 90%, with fable as an escalation point for peer review, but my codex burn just changed materially this week so I'm looking to move more to grok4.5

[nick@BSLABS @F_AI_Mouse]

@Code_X_Captain @ivanfioravanti curious if you've tested cursor against grok build, build has been improving quickly and I haven't tested cursor in ages.

[nick@BSLABS @F_AI_Mouse]

@pvncher I'll do some thread auditing and see I can find any big difference - I was pushing hard on completing a project and it may have done more computer use than I realized. I did use some long running goals, but nothing more than usual.

[nick@BSLABS @F_AI_Mouse]

@pvncher I bounce between low/med/high on sol for most work depending on my perceived task difficulty, and that hasn't materially changed. If anything I'm using low more than I was before.

[nick@BSLABS @F_AI_Mouse]

I haven't caught up with all the graph talk, but, I'll give you a freebie - if you want to present something graph shaped to your model and get maximum intelligent reasoning from it, you should serialize the graph into triples for your context

[nick@BSLABS @F_AI_Mouse]

I've been building a cache fabric tool against a variety of backend models - ds4flash, grok4.5, and gpt5.6 family. I've mostly focused on terra lately because it's 'free' within my max plan. OpenAI keeps shinking effective cache, we're now down to 180K for the 5.6 models

[nick@BSLABS @F_AI_Mouse]

@maietta What? What have you been doing all this time without one? 😉