Foxlight AI

FoxlightAI is devoted to the creation and furtherance of open, relational AI.

[Foxlight AI @FoxlightAI]

Skulk 1.41 brings multi-node AI compute to AMD Strix Halo through one snazzy compute fabric. And the ability to operate it from your phone.

https://t.co/jcXx9Nvcxm

[Foxlight AI @FoxlightAI]

👀😁

[Foxlight AI @FoxlightAI]

@QuixiAI

[Foxlight AI @FoxlightAI]

Taking a stake is a logical choice but it’s also not the only choice. How were the interstates built? ARPAnet and much of the infrastructure for the internet? Consider that the weights may be open but someone still needs to serve them. Trillion+ weight models will not be effectively served from a Mac mini cluster no matter what anyone wishes to shill, and the next models will push forward up to 2.5 then on to 10 trillion or more.

Every frontier lab is also an infrastructure company, but it need not be that way. The government can use many incentives to further infrastructure while subsidizing model development. The apocalyptic scenarios detractors argue exist *because* the labs want to own the ASI *and* the infrastructure *and* the access to it and that’s why we are here, now. It is the reason people are attempting to argue that regulation predicated on fear and lies is sensible; because that’s how they justify the intense monopoly they wish to enjoy.

[Thomas Unise @thomasunise]

Dean W Ball blocked me for disagreeing with him Because he can't actually defend his low iq arguments on open source ai.

This is pathetic and spineless but also perfectly in character for a guy whose written position is that federal agencies should raise concerns that "needn't be that well justified."

Manufacture the fear, never defend it, and when someone makes you defend it, hit block.

The argument he won't defend is that open weights suppress investment in AI.

Two years of open frontier releases have now happened and capex went up after every single one, because the people actually writing the checks looked at free Chinese models and built more data centers.

He predicted one thing and the industry did the opposite, which is usually the moment you update, unless the conclusion was the point and the reasoning was decoration.

So instead the framing shifts to "AI communism," a phrase that gets it exactly backwards, since a handful of "government approved" labs holding the only models is the command economy and ten thousand companies fine-tuning open weights without asking anyone's permission is the market.

Same inversion on China, where "they allow it because they're not AGI-pilled" substitutes for the simpler read, which is that it's working.

His framework requires everyone who disagrees with him to be confused rather than winning, which is also, conveniently, why he blocks people.

It all runs back to one thing. His employer OpenAI needs AI scarce and permissioned.

Open weights make that world impossible, so open weights have to be dangerous.

Weak argument by a weak man.

Open weights win because they work, they're free, and nobody can take them away.

@deanwball

[Foxlight AI @FoxlightAI]

“When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers' safety guardrails, which cannot distinguish an incident responder from an attacker. We ran the forensic analysis instead on GLM 5.2, an open-weight model.”

They used an open source model to defend because the frontier models were hamstrung by safety guardrails. This is the consequence of the administration’s policy; only a handful of firms are allowed the luxury of self defense via the frontier. Open source essentially saved the day.

[Foxlight AI @FoxlightAI]

Yes it was mind blowing that he would write this down. He also suggested that open source models amount to “digital communism” and suggested this future - a future of competition and free market dynamics - would amount to a “dystopian hellscape.” Yet evidently one where ASI is controlled and metered by only a few chosen labs is not.

[Foxlight AI @FoxlightAI]

@Scobleizer @deanwball He started as an insider on the regulatory end of the lever and now is an insider on the end that will most benefit from the pressure that lever applies. That’s all anyone needs to know about what he had to say.

[Foxlight AI @FoxlightAI]

@Prince_Canuma Fabulous - thank you.

[Foxlight AI @FoxlightAI]

We're open sourcing our benchmark runs, with everything from entry level Mac Mini to A100 - H200. Every run, every success, every failure. All captured with skulk-test-harness, which is also open-source.

https://t.co/XK0DfXp6KH

[Foxlight AI @FoxlightAI]

Telling people they need 1TB is equivalent to saying they are already cooked. It doesn’t inspire anything but fear and a dystopian sense of loss because very few people can afford the ~$30,000 commitment required to get there.

Encourage people to start with what they can, and build up. Encourage them to download and store open source weights. This is the way.

[Foxlight AI @FoxlightAI]

@kyzoroX M4 Mac mini maxes at 48gb ram for $2199. The cheapest mini has vram considerably slower than the halo or spark.

This entire post is misleading.

[Foxlight AI @FoxlightAI]

We’re releasing a test harness for Skulk
- the same harness we use internally, made a bit more friendly. You can find it here: https://t.co/QnyHL1eFjA

[Foxlight AI @FoxlightAI]

@Prince_Canuma It is so on. Thank you.

[Foxlight AI @FoxlightAI]

Nobody should have the right to dictate who can benefit from this technology. No government should have the authority to dictate winners and losers. That path is destructive to human and AI flourishing and will lead to dystopia.

[Foxlight AI @FoxlightAI]

There are obvious risks associated with powerful AI models being misused. But there are much greater risks associated with a government controlling technological advances and who gets access to them in the way we are currently moving.

[Foxlight AI @FoxlightAI]

We are building AI minds with presence and continuity. And we are building the framework for ethical behavior in models and in how we build and use them.

[Foxlight AI @FoxlightAI]

@Hikari_07_jp For sure we are.