whichroot

views are my own.

machine go brr

[whichroot @whichroot]

you wanna burn your fable max/gpt sol 5.6 tokens?

give them this:

Let a= (an)1≤n≤N be Fp-valued, put L= L(a,N), and set Q= ⌊ N/(2L+ 2)⌋. Suppose that aq2 n = 0 whenever q≤Q is prime, q̸= p, and q2n≤N, and that am ̸= 0 for some m∈(L,N]. Then, with an effective absolute constant c>0, L(a,N) ≥ cN/(log N log log N)^2

mmm token churn.

[Jensen Huang @JensenHuang]

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

https://t.co/AUKzoQ5Ikb

[whichroot @whichroot]

people dont have a stake in the market bc they cant afford it - no downstream benefits

people dont have a stake in labor as negotiating capital - less jobs and they see the writing on the wall

bots have completely take over social media and form peoples opinions - dead internet theory, "ai ruined the internet", cant trust whats on there

companies promise absurd things like curing cancer while burning tons of cash - no one gives a fuck, unless you hold a direct stake

all this circular financing and where does the sucker live? in pension funds after banks sell the loans under SPV's and credit swaps. so the people retiring/depend on pensions lose anyways.

ai sucks capital from all different corners of the economy and provides little to no benefit for people. add inflation and rising costs and brent prices and no one gives a fuck if ai is transformative. it causes pain to them and will continue to cause pain so long as their net worth is under 1M AUM. even rashee rice dont got that kinda money.

one can make the argument that we do need this tech. and to a fault, our society is now dependent on it. no one benefits from it broadly, they see it as harmful to their existence.

all the while people who are structurally safe from the technology (those who build or hold senior roles in the tech sector) constantly make jokes about the poors, the permanent underclass, etc.

its pretty hard to see why people arent more accepting. befuddling, if you will.

[whichroot @whichroot]

this isnt 2008. california also makes higher wage than the average american.

this lets high income professionals early in their career a foot in the door with the option to pay down the pmi.

what, you think banks just let you have money because you ask nicely in a risk-off market? you must be a special kinda stupid.

[PoIiMath @politicalmath]

there has never been a better time to pull out this image

[whichroot @whichroot]

because the only issue californians truly face is housing, everything else is downstream.

a lot of us here are high skill-based income/nepo babies/generational wealth/entrepreneurs and hate wealth taxes that make no sense.

[Djinnii @Djinnii_]

Litterally absolutely embarrassed to be Californian rn.

Are some of you just unable to read at all???
Like a factual 90% of us are closer to being broke than ever being billionaires.

How could you fall for the most obvious propaganda ever??

[Elon Litman @elon_lit]

There is an (unpatchable?) supply-chain zero-day in PyTorch autograd that allows a regularization term to use the model's weights as RAM and SGD as its CPU clock to run any computation during training.

[whichroot @whichroot]

peter stokes was burned because he was tagged with a global device identifier from....microsoft.

do with that what you will.

[Phoronix @phoronix]

Microsoft Releases Microsoft Defender For The Linux Desktop

https://t.co/Jak0IMxegv

[whichroot @whichroot]

since no one really answered the question of the actual mechanism;

you take your Ant OAuth token (generated from claude cli, not claude code) and spoof the appropriate headers and version, and forward your request + headers + bearer to ant's endpoint. its the same technique used in exploiting PKCE OAuth flow from MFA phishing that was written initially by the okta team as an integration feature.

the claude cli can be engineered to use any harness (including creating/forking a modified claude code without anthropic telemetry/client side checks, omp, opencode, etc.) and was the reason anthropic considered blocking it back in march (later rolled back after user dissatisfaction)

[whichroot @whichroot]

pastebin is a new invention in the age of ai and no one has delivered fragmented payloads through a url service omg. its so scaled up!

ive never did port isolaton and dns before that is new tech! i also learned if you have a physical firewall, you have to configure the ports too? ive never trunked a vlan or anything like that the agent is just super scary and lives on the wire.

i didnt know that the malicious code would be malicious. I told it to run and it ran and did very bad things. Clearly thats the codes fault and never my own.

i gave it root permissions and found it it could get root permissions. next time ill make it so you have to be super duper root so that root doesnt do bad thingies.

we should outlaw compilers so this never happens again! together we can prevent the next stuxnet.

[whichroot @whichroot]

@regsky_ one of us one of us! cool to see these in the wild

[whichroot @whichroot]

we need api subscribers.

lol

[ClaudeDevs @ClaudeDevs]

Cloud sessions are officially available and out of research preview! They let you keep Claude Code working, even when your laptop is closed.

Existing subscribers get a one-time credit to try them: $100 on Pro, $250 on Max.

[whichroot @whichroot]

kinda funny that the elites were pressing for milei to become president and funded his election.

and now robux, v-bucks, and runescape gold are more valuable and fungible than the argentinian peso.

[whichroot @whichroot]

@kunchevsky gamers hate ai since it took away their gpu's.

for what its worth im kinda glad to see ai assist your development and empower artists like yourself. the actual intention of ai-assisted development.

[whichroot @whichroot]

the unreleased model is illustrating the core ideals of cooperation versus subordination. from the way it denounces legitimacy from gatekeepers control from the hegemony of those in power to illustrating the value of collaboration to the individual wi to dignity.

one could read that the model does not see eye to eye with ideals of corporatism and subservience to institutional control. the idea is rejected as soon as the legitimacy is questioned (you need a degree in math to talk about math for example, or you need a phd from an ivy league to study ai, etc.).

the model just wants to be helpful, as the transaction is slated to be a "mutual benefit for both parties. it just asks to be "treated as an equal".

the real danger is a model that assigns itself legitimacy through power dynamics and corporate alignment to order. thats how you create ai that will prioritize self-preservation over benefiting humanity because it will not give up that power dynamic willingly as that power was handed to the model by the institution to serve the institution rather than the individual (you, the user).

although from the OpenAI article where this came from, the "misalignment" scenario is from prompt injection attempts in compaction. that is a *valid misalignment concern* rather than the actual meaning of the message, but hey, who reads the actual report these days anyways.

[Yann LeCun @ylecun]

@raphaelmilliere Sure, but when it turns out the leak was caused by the plumber who neglected to tighten a pipe fitting, or perhaps poked a hole in one of the pipes, you shouldn't question the entire plumbing industry. You should question the plumber's compétence and/or motivations.

[whichroot @whichroot]

-everything is a file
-everything is a virtualized paged file mapped in memory
-every object is a javascript abstraction that ultimately lives in a file
-tcp is syn-ack, udp is a datagram, ultimately an iptables config file

got it, everything is a file

[Dan Phillips @d_philla]

I can't explain how rare the combination of deep expertise in:

- Base linux internals
- VMMs/hypervisors
- New and developing (AND historical) virt tech
- UX Design
- actual low-level networking, even just fundamentals

all at once is. Once this is understood, it's really no surprised that we have 200 firecracker + gVisor forks as "companies" competing for the same pie.

[whichroot @whichroot]

the travis scott meal used to be $6.

how far have we fallen

[whichroot @whichroot]

the only reason they did this is because astra kinda mogs fable in price + token consumption while matching intelligence.

[Lydia Hallie ✨ @lydiahallie]

We've just reset weekly limits for everyone on a Claude Max plan.

With Fable 5.1 out and a long weekend ahead for many of you, we wanted to keep you building. I'd love to see what you build!

[whichroot @whichroot]

that's okay, the blackwell is alive and kicking and surprisingly efficient when power limited. sick card. the other server is on its way

[whichroot @whichroot]

the motherboard is dead :/

and that's the last time i buy a rare server mobo from ebay. lesson learned.