iiinova — where your digital self keeps evolving

Can also report on Mistral 4. I've been running it for some tasks on my laptop direct from Mistral, and, at just under 100tps and the price they're asking, I am on the "meh" side. I still think its an important model, but I'll be interested in seeing what comes AFTER this one.

000

Bonsai 27b 2 vs. Qwen 3.8 Next Flash IQ1 (Strata) An LLM entirely in GPU vs. one spread over GPU, RAM, SSD and even the processor doing some worok. This is my custom benchmark, so of course YMMV. Honestly, not too revealing. I don't think I'll be able to say how good Strata is until I just use it.

000

Mistral 4 might not be leading at anything... but it's still an important entry into the open weight space. I signed up for a Mistral API to run it through some testing. The AI world is better with Mistral in it. If Mistral 4 is "good enough" and, relatively fast, it could be something I use.

2140

Let me get back to you :) I am going to a/b it against Bonsai 27b 2.

Successfully got n³ setup last night. (NixOS + Niri + Noctalia) This was by far and away the most troublesome Linux distro to set up for the first time. That said, I can see why it'll be good in the long run. Future installs should be significantly easier after learning some hard lessons.

020

Okay. This is... I can't think of a better word than "interesting." A $3500 computer with [hidden] ports, that appears to be designed to be hardware AND software personal AI solution. I'm curious what software it includes, the website is a bit shallow on it. ghost.ai

ghost.ai

000

I will say, it does appear to be *very* stingy on tokens though, which is a delightfully refreshing change of pace.

090

Beam is here. Reflection AI's new 501b MOE model. A promising addition to the open weights space from an American lab. The numbers alone aren't enough to make me want to switch from GLM 5.3 Flash, so we'll have to see how it actually performs. reflection.ai/blog/introdu...

2170

Yep. One of the best casting jobs in recent history. I *loved* this season of Lanterns, and it was almost entirely because of Kyle Chandler. Coach Taylor is undeniable.

Well, this is a new record...

000

Update Sunday on the homelab: One major server upgrade (Ubuntu 24->26), 25 containers updated across three servers, two major version jumps, one dead service decommissioned, everything verified and zero rollbacks. Slow, steady, staged upgrades with backups before every step. Nothing caught fire. 💪

110

I've been wondering what they're up to. Feels like they fell off the earth. I hate what they've done to their token plan, but will be nice to have another contender on the field again.

000

I spoke too soon. The victory I thought I had late last night getting @nixos.org set up ended in tears as I somehow, managed to lock myself out. 🤦‍♂️

000

A few things of note from this, that shows what kind of day its been: - Finished a months-long TUI project earlier today. yay. - Mercury crossed 1000 tps for me for the first time. Yay! - LFM2.5 is my local background/memory model. If it's getting ANY cache at all, that's the sign of a bad day.

000

Turns out, I suck at giving up. Banged my head against the keyboard for another two hours and got @nixos.org up and running. I'm sure I'll be happy about it in time, especially when I'm migrating other systems to Nix. But right now all I can think about is NO MORE DISTRO-HOPPING!!!

000

Joshua White @jrw14.whnc.me · 03/10/2026

Blech. First interactions with NixOS have *not* been good. Having a rough time getting it set up with LUKS on my X1 Carbon laptop. Immediately going into a "root account locked" on initial startup. Don't feel like fighting it anymore tonight. If Linux was easy, everyone would do it? 😵‍💫

000

Wiring up a custom e-ink dashboard for the homelab. 🤓

- Does not spend more than $30-$50 a month on AI, and has not found anything limiting my ability to complete a task despite that? - Spends more time thinking about how/when to get more inference LOCAL vs. whatever the "next big thing" is? - And yes, thinks torturing models is bad, 🙃😕

000

- Chronically experiments with new models, use Claude/Gemini at work, and would be a-okay if intelligence stopped at Qwen 3.8 27b level? - That said, my daily driver is GLM 5.3 Flash and <loves> it - Is no longer impressed by intelligence gains? Impress me with efficiency and speed, not AA scores.

100

mr. TIM @timkellogg.me · 02/10/2026

Prefaced with "I am not an expert, maybe I'm wrong..." but here I am, someone who spends 5-7 billion tokens a month, wondering the following things: - Am I the only one who keeps context under 225k, always, regardless of model's limits? - Thinks being locked to one AI provider is a terrible idea?

100

Reposted by Joshua White

i get annoyed by anti-DC discourse, but this is a very very very good outcome, if this becomes normal www.linkedin.com/pulse/race-o...

America can lead the most important race of our generation, and getting the facts right is a big part of how we do it. Today we're sharing our Amazon Data Center Commitment, a clear set of promises for how we build and operate, and launching Built Together, more than $1 billion over five years for the communities where we build. I wrote about what's at stake and why this moment matters.

3221

I've only been "settled" on EndeavourOS for a couple months, but seeing all the posts (and my own research on) NixOS has me seriously considering (yet another) distro change. Plus... something kind of appealing about n-cubed (Nix, Niri, Noctalia). Listen, I'm easy AND I'm fickle, okay?!

NVIDIA has some good offerings. And, I think with their acqui-hire of Poolside, they'll get even better. Poolside's 2.1 models were among the best in their size before they shut down and moved to NVIDIA. Guessing their talent will make a big impact.

Joshua White @jrw14.whnc.me · 01/10/2026

lmao, that tracks, but then I'm even more confused about when to expect my pre-order!

101

Joshua White @jrw14.whnc.me · 01/10/2026

I am slightly confused by "Round" and "Batch." I appear to be in round 2, batch 3... what does that mean for me?

100

Joshua White @jrw14.whnc.me · 01/10/2026

For what it's worth, thanks for the love on this. Bluesky is a pretty neat place. Also, just an additional note... these two images represent a reduction in shirt size from 6XL to M. 🤯

000

I have a subscription to Hyper from @charm.land

okay, you're kinda on fire today, lol

060

I take a lot of pride in being open to being wrong, not stubborn, and willing to engage with new things. That said, I am _absolutely not_ calling it "SI."

060

I recently hit a new personal best at the scale, which made me go digging through old photos. I'd say I barely recognize who I was four years ago, but body-dysmorphia is a real pain in the ass, and the first pic remains all I see. Still, losing 256 pounds has been absolutely life changing.

4570

Wait, y'all have degrees of any kind?! ...high school dropout here

110

North Carolina Courage @nccourage.com · 29/09/2026

I pay $20 a month for (what feels like) unlimited GLM 5.3 Flash at 220 tok/s. Again I ask... what are y'all doing?!

110

Reposted by Joshua White

Ash answered the call 🇺🇸

Ashley Sanchez holds a USA flag "Called Up USWNT International Friendlies"

210420

I see you, Ashley Sanchez!!!!!!!!! Hell yeah! @nccourage.com

011

One thing I still, to this day, do not quite understand... is what kind of work everyone else is doing who really think they *need* Opus/Astra level of AI? Hell, even Luna/Sonnet level. I'm just over here happy as a puppy with a new toy with GLM 5.3 Flash and DSV4.1 Flash, still.

000

Finally got around to getting Excalidraw set up on the server. There's something quietly endearing about prompting an agent in Hermes with very little ("a drawing inspired by the movie Interstellar") and just seeing what happens.

The day after Muse shared users text messages? I don't get the people who would sign up for this.