Yuri

i like my functions smooth and my life discontinuous

[Yuri @Yuure96]

@LingoraAI just followed, i think this is one of the more underrated problems in ai

[Yuri @Yuure96]

@chloxjin this is gonna be stuck in my head now

[Yuri @Yuure96]

@AnnieBot4573 ai is changing creative work, but it doesn't make every artist interchangeable

[Yuri @Yuure96]

@Alexvx_nft privacy and lower costs together are a pretty compelling combo

[Yuri @Yuure96]

@MaxScore interested to see how manako defines a new benchmark here

[Yuri @Yuure96]

@JohnTzimiskes97 @Squattocrat curious what kind of society you think is actually realistic in the long run

[Aurex @0xAurexx]

nvidia wants $10,000 for one gpu. intel just showed up with four cards for $4,000 that sit right behind it.

the intel arc pro b70:

→ 32gb of vram per card. $1,000 each
→ 4-way config = 128gb of vram for the price of a used car, not a new one
→ doubles the performance of nvidia's dgx spark. or more
→ full workstation build lands a bit over $4,000

the lineup right now:

spark - compact, ai-only focus, and four b70s run 2x+ past it.
rtx pro 6000 - still the king. $10,000. one card.
4x b70 - the middle ground nobody was selling until now.

the catch nobody posts: four cards don't equal one big one. you're splitting models across gpus, dealing with 4-way setups, and the $10k nvidia still wins raw performance. this is the value play, not the crown.

but 128gb of vram for $4,000 didn't exist a year ago at all.

and unlike the spark, these also game.

[Mura @0x_mura]

🚨Netflix spends over $100,000,000 a year dubbing its catalog into 30+ languages.

a 21-year-old spent $46 last month and made $6,400 dubbing other people's viral videos into languages nobody else touched.

> Claude translates the hook and finds the gap: 10 min
> ElevenLabs clones the voice and dubs the video: 15 min
> CapCut adds native captions: 10 min
> Make publishes and tracks the view count: automated

$46/month in tools. $6,400/month back. 0 original ideas, 1 language nobody else dubbed into yet.

netflix hires translators by the hundred. he hired 3 tools and a search bar.

exact settings are in the article below.

[Yuri @Yuure96]

@JaroslavBlek3 thanks for sharing, i'm going to read more about this

[Yuri @Yuure96]

@yewfnet just followed, the mac mini video is the one i'm waiting for

[Yuri @Yuure96]

@TruthNetwork24 @CVSHealth @cvspharmacy @CVSCareers @CVSHealthJobs curious if cvs has responded to any of these concerns yet

[seeco @seeconvm]

Claude + Obsidian + n8n + massive local storage turned one desk into a private AI workspace.

No scattered cloud tabs.

No monthly storage anxiety.

Everything stays local.

New files get ingested automatically.

Research gets summarized.

Projects compound instead of disappearing into folders.

The interesting part isn't the 316TB.

It's ownership.

Most people rent their memory and workflows from cloud platforms.

A small group is building infrastructure they completely control.

The second brain stops being a chatbot.

It becomes an operating system for knowledge.

Bookmark this before local AI workspaces become normal.

[Yuri @Yuure96]

@HouseofElon @CENTCOM a lot of this looked impossible just a few years ago

[Yuri @Yuure96]

@LeeLinAI123 the jump in quality over the last year has been kind of crazy

[Superior @andreysuperior]

Everyone is talking about AI assistants that read your calendar and wake you up with the weather.

Nobody is talking about what that assistant is actually doing behind the scenes.

It's not a gadget. It's an operations department. Scheduling, lead scoring, invoicing, reporting, all handled before the coffee finishes brewing.

A company doing that manually pays a 10 person team $50,000 a month. The automated version costs $200.

He watched the demo and skipped the part everyone else stops at, the part where you say thanks and go back to your day.

He built five more workflows, packaged all seven, and sold the system to other founders.

$2,500 a month per client. Eight clients. $20,000 a month, recurring, from something he built once.

Everyone else saw a smart assistant.

Nobody noticed it was a product.

[Grimmer @0xGrimmer_]

4 MAC STUDIOS STACKED ON A DESK POOL INTO 2TB OF MEMORY AND RUN A TRILLION-PARAMETER MODEL LOCALLY.

that clip is a build: four mac studios linked into one cluster, running kimi k2 (a trillion-parameter model) at ~28.6 tokens a second, entirely on a desk.

the trick is apple's unified memory. four 512gb mac studios pool into ~2tb, enough to load models a $2,000 gpu can't even open.

what it runs: kimi k2, deepseek, llama 405b, the biggest open models on earth, at real chat speed, nothing leaving the room.

the dashboard: four nodes, around 39°c, pulling 17 to 22 watts each. a "supercomputer" quieter and cooler than one gaming pc.

now the honest part, because this is not a budget hack:

a 512gb mac studio runs ~$8,000 to $9,500

four of them is a $30k to $40k rig, not a $600 mini

and apple has since pulled the 512gb option, so this exact build is hard to buy new

so what are you actually paying for? frontier-class models you own outright: no rate limits, no per-token bill, no data leaving your house, ever. that is the thing this replaces.

the uncomfortable part: "supercomputer" used to mean a room and a utility contract. now it is four boxes on a shelf and a wall outlet. the capability reached the desk faster than the price came down.

no datacenter, no rack, no cloud provider holding your model.

save this for the day the $40k rig costs $4 k.

[CyrilXBT @cyrilXBT]

PEOPLE ARE NOW RUNNING PORTABLE OPENCLAW RIGS ON BATTERY POWER TO CATCH LIVE EVENT CONTENT ON THE SPOT.

A Mac Mini running OpenClaw, off a high output USB-C battery bank, mobile and untethered from any outlet.

The advantage is timing. Most creators wait until they are back at a hotel to process footage and write up takeaways. By then the thread is stale and someone else already posted it.

A portable setup catches keynote audio live, extracts what matters, and gets it out while the room is still listening.

Mac Mini plus OpenClaw plus a battery bank rated for its power draw is a real, documented setup.

The workflow this description points to is genuinely one people are building right now.

[Yuri @Yuure96]

@MoonPieJoe lol

[Yuri @Yuure96]

@Mrsolexpert_1 clean character design. that's usually a good start for meme projects.

[Libriscent @libriscent]

I have two expectations of people. TWO.

1. If you’re interested? Fucking act like it. Consistently.

2. If you fuck up? Own it and fucking fix it.

The bar is in the sub-basement of hell, guys. Come on.