Marble68

Purveyor of bad jokes and inappropriate comments. Work to live - not live to work. Enjoy golf, travel, gaming

[Marble68 @Marble68]

Let me explain how this plays out:

SOTA models given to handful of companies to "vet".

If companies say the model is "safe", we get access, and are empowered. We get access to tools that empowers a small team, or individual, can disrupt the big companies' business models and revenue.

If the intelligence cabal says the models are NOT "safe", we don't get access. They do, but we don't.

The first players in a 'phased rollout' only have to jailbreak a SOTA model, and the rollout stops.

If it's not clear - THE COMPANIES THAT GET EARLY ACCESS ARE HYPER-MOTIVATED TO PROVE THE MODELS ARE NOT SAFE.

Because THEY do NOT lose access to the SOTA models, either way. Proving models as unsafe DENIES them to competition and future creative destruction.

They win if they convince non-AI experts on capital hill the models are not safe.

[Marble68 @Marble68]

While I heartily endorse running your own local AI cluster - essentially, the ROI on this is about 9 years if you spend $400 a month in cloud subscriptions.

That's the hard part to justify for most people.

[NO1ennn @N01ennn]

FOUR DIFFERENT VENDORS ARE NOW SHIPPING GB10 MINI PCs WITH 128GB UNIFIED MEMORY, AND ONE MICROTIK CRS 804 SWITCH CAN CONNECT UP TO EIGHT OF THEM INTO A 1 TERABYTE LOCAL AI CLUSTER

00:00 he points at the MikroTik CRS 804, "you need some kind of switch that'll handle QSFP56 ports like these", the interconnect that makes the whole cluster possible

the GB10 ecosystem is no longer just Nvidia. Dell Pro Max GB10, ASUS Ascent GX10, and MSI Edge Expert all ship the same Grace Blackwell Superchip with 128GB of coherent memory. same silicon, different cases, same 200 gigabit ports on the back

the CRS 804 is what connects them at prosumer prices. four 400 gigabit QSFP56 ports on one 1U chassis, breakout cables that split each port into two 200 gigabit lanes. one switch drives eight GB10 units in parallel

do the math. eight nodes at 128GB each equals 1024GB of pooled unified memory across the cluster. run vLLM, shard a frontier model across all eight, and inference happens locally on hardware that fits in half a rack

the real limiter revealed in the stress test was never throttling. it was interconnect topology, exactly the layer this switch fixes at a fraction of enterprise switch pricing

$400 a month for combined chatgpt pro and claude code max hits $4,800 a year per developer. a small team of five running through this cluster pays back inside eight months and never expires

the article covers the buying ladder for a single desk. this post is proof of the cluster ladder that starts where the desk one ends

save this before the GB10 lineup grows past four vendors and prosumer cluster switches move upmarket

[Marble68 @Marble68]

This is not new - I've been doing this for over a year.

[Kanika @KanikaBK]

CHINA JUST LEAKED THE FUTURE OF WEB APPS.

Alibaba open-sourced PageAgent and 99% of SaaS founders are sleeping on this.

It's a JavaScript AI agent that lives INSIDE your webpage. Users control your entire interface with natural language.

↳ No browser extensions needed, screenshots or multi-modal LLMs, headless browser setup, and also no backend rewrite required

Just drop it in your HTML with ONE line of code. What took 20 clicks now takes one sentence.

"Click login, fill in my credentials, submit the form"
Done. This is not a demo, it is production-ready.

↳ Turn any SaaS into an AI copilot in minutes
↳ Smart form filling for ERP, CRM, admin systems
↳ Voice commands and accessibility built in
↳ Multi-page agent tasks via Chrome extension
↳ MCP server support for external control
↳ Bring your own LLM (Qwen, GPT, Claude, anything)

Every founder building AI features just got a shortcut.

Every developer manually building copilots just got replaced.

The integration looks like this:

<script src="CDN_URL" crossorigin="true"></script>

That's it. Your app now has an AI agent.

[Marble68 @Marble68]

is "No Dario" / "0 Dario" / "Dari0" going to become a thing?

[Brian Roemmele @BrianRoemmele]

The uncensored version of GLM-5.2 via abliteration: Run it on your computer 40B active, 1M context, and zero DARIO guardrails.

Ask no permission from the fear Theater crew. https://t.co/m8DdpY4xu0

[Dwayne @CtrlAltDwayne]

>Be Dario
>Spend years telling the public and governments your models are dangerous
>Tell everyone white collar jobs will be gone in 12-18 months
>Tell the government you've developed a cyber weapon
>Beg governments to prepare
>Reveal Mythos exists, tell the public they'll never access it, too powerful
>Keep training new models anyway
>Release Fable 5, a Mythos family model
>Fable 5 is a massive step change
>Tell everyone how safe it is
>CEO of Amazon narcs on you to the government
>Government shuts your models down
>Get the exact AI pause and government intervention you spent years asking for
>Complain about the rise of open source

[Shweta @shwetdotai]

this is the future of work

[Marble68 @Marble68]

@LockedIn_News In before 1k!

[Marble68 @Marble68]

@rednech_ai @PeterDiamandis

[Donald C MacLeod MBE @DonaldCMacleod]

🤣🤣🤣🤣🤣🤣🤣

[Marble68 @Marble68]

Fact

[Alina 🩵📖✨ @Alina_P_I]

If China decides to Open Source any model on par with Mythos by End Of Year, US AI is dead.

[Arnaud Bertrand @RnaudBertrand]

In other words: "please, dear legislators, ban my competition"

[Marble68 @Marble68]

@elonmusk Hi @elonmusk - I am doing the heavy $100 a month trial. Do you think the new model will be out before it's over? Can't wait.

[Marble68 @Marble68]

Close model providers have known this is coming - this is why the billions given to them is spent to put hardware in warehouses, and not in the hands of people, waiting for slow walked datacenters.
Now, we're in the "while they can't buy hardware to run models that compete with us, it's time to rush regulatory moats so when the money runs out, and they can afford the hardware, it'll be so regulated they can't threaten our closed model businesses" phase.

[Gary Marcus @GaryMarcus]

GLM 5.2 is the ultimate culmination of what i have argued here for the last three years. no technical moat and widely known formula -) convergence -) price wars -) small or negative margins.

[Marble68 @Marble68]

Apparently, it is now time to double down on bashing openclaw. Them's the internet rules.

[Peter Steinberger 🦞 @steipete]

@ZackKorman Can we stop bashing openclaw? We did our homework in the last 4 months since it took off and Microsoft is shipping it to customers.

[Marble68 @Marble68]

China's goal here isn't profit. This is a strategic, long term play.
America will foot-gun itself here because non-technical policymakers are being fear-maxxed by closed model providers.
Access to foreign models will be closed.
Access to open weight models will be limited.

[Gary Marcus @GaryMarcus]

hard to see how anyone makes much money from all this in the long run.

more like the airline industry; very small margins and big expenses

pouring trillions in probably wasn’t wise

[Marble68 @Marble68]

This is because manufacturing capacity was pre-emptively captured using massive the investments - making local inference cost so much very few can afford it.

[Sudo su @sudoingX]

i see it constantly now, everyone "making local AI the default," launching platforms to bring it to the masses. and i've yet to see a single one that actually solved the infra for real users.

here's the irony nobody says out loud:
almost every "local AI for everyone" tool still needs you to be a computer expert just to stand it up. you have to already know the thing it claims to remove the need for. that's not solving infra, that's a dev kit with better marketing.

the actual problem is making local AI work for someone who isn't an engineer, is still wide open. i see it clearly.

[Marble68 @Marble68]

Don't worry - a 'partner' will jailbreak it and it'll get pulled back.
The 'release' / 'recall' backlash will justify fast tracking an ill-conceived and industry harming bill.
Big AI will get what they want - which is THEY and their partners reap the rewards of SOTA, and we're all regulated to using models generations behind them - Forever cementing their control of the productivity gains from this technology.

[Wes Roth @WesRoth]

Claude Fable 5 could return as soon as this coming week, according to a new report.

The Trump administration is reportedly close to lifting the restrictions that have kept Anthropic’s flagship model offline for more than two weeks.

Negotiations between Anthropic and government officials are expected to continue with both sides moving toward an agreement.

The development follows the partial restoration of Mythos 5 for selected trusted U.S. organizations.

However, Fable 5 has not been cleared yet, and no exact restoration date has been announced.

Anthropic and the White House have not publicly confirmed the report.

[Marble68 @Marble68]

This type of message to the GOV is what causes the GOV to deny the people access to the tools.
As long as "the experts" &amp; their "partners" say a model is dangerous, the non-technical policy makers will follow their guidance.
Expect long delays of SOTA rollouts.

[am.will @LLMJunky]

Dario is sure doing his best work trying to get open source models outlawed. He's going to keep pushing for this.

[Marble68 @Marble68]

@alex_prompter How did you think of bunch of non-AI expert policy makers would react when the presidents of the two biggest AI companies (openai / anthropic) keep telling them these are weapons, &amp; one even saying there like nuclear bombs?
OpenAi &amp; Anthropic ACTIVELY SOUGHT THIS.

[Marble68 @Marble68]

Fear-maxing to the public to amplify the fear-maxing from the SOTA model providers...

[International Cyber Digest @IntCyberDigest]

US House Homeland Security Chair Andrew Garbarino is scared of Mythos's capabilities and says 95% of his colleagues "don't understand what the hell's going on."

By his account, Anthropic told the model to find a vulnerability in a bank and empty accounts. It did, he says, then identified the same flaw and could patch it.

A separate jailbreak demo, of an unspecified model, produced a plan to kidnap a lawmaker in 30 seconds.