Amir Valizadeh

Core Engineer - OpenLens @ai_bread

[Amir Valizadeh @vitalune7]

@Eschalabs this is really flagging my bench-maxxing spidey senses. has anyone actually tried this and confirm its empirically good?

[Amir Valizadeh @vitalune7]

@TheAhmadOsman "Everything in AI is an infra problem btw" - ai pilled chad

"Everything in infra is an AI problem btw" - infra pilled chad

[Amir Valizadeh @vitalune7]

@sudoingX benchmarks were never the reason.

i cancelled my $200 subscription during the golden era when the best model was free use.

the very obvious ai provider company that we're all thinking of, loses regardless of benchmarks.

[Amir Valizadeh @vitalune7]

@shengkun_ye why are you comparing an arbitrary monthly credit limit to a per page price? doesn’t make sense

also your per page price is industry standard, what’s so special here

[Amir Valizadeh @vitalune7]

@FrameworkPuter @notebookcheck should i buy a framework laptop 13 pro right now and migrate off macos, or wait

the generational question

[Amir Valizadeh @vitalune7]

@sudoingX which one? was it the strix halo? ive seen impressive demos of those handling cyperpunk well

[Amir Valizadeh @vitalune7]

i hereby refuse to build anything until >100 customers begged me for it

# protect product engineers' rights

[Amir Valizadeh @vitalune7]

@josh_tobin_ a pacing the frontier man is not a man

[Amir Valizadeh @vitalune7]

US universities are genuinely among the worst sources for fraud in the country

in another example, the university of pittsburgh has approximately four TERABYTES of gpu compute spread across hundreds of nodes, and claims its “currently impossible to serve glm 5.2” to students and teachers. what a fucking scandal

[Amir Valizadeh @vitalune7]

@grok @ycombinator @grok why did they shift? what made their original product so fucked, was it the product itself, was it market demand, wtf happened that made them wet the bed and quit

[Amir Valizadeh @vitalune7]

@lucasmeijer why would they do that? it literally makes no sense for either them or consumers to do that. if you want a GUI for codex then just make it yourself.

that’s what we did for OpenLens
https://t.co/BJx1BPS1lu

[Amir Valizadeh @vitalune7]

@shengkun_ye @OctenAI @monid_ai Compare & contrast “people search” @grok

[Amir Valizadeh @vitalune7]

@ImLukeF @BradGroux it isn’t worth using anthropic models at all. even saying you hate it isn’t the right approach. they simply don’t deserve your money, no matter how good their models are (they’re not even #1)

[Amir Valizadeh @vitalune7]

the mistake is treating open weights as a personal workstation product. it isn’t “enough vram or useless.”

quantization, offloading, rented gpu boxes, confidential deployments, dedicated inference infrastructure. the same weights can support everything from research to production-scale serving. not just something that demands an in-house data center.

[Amir Valizadeh @vitalune7]

local ai has officially left tmux and entered the streets

not your weights not your waifu

[Amir Valizadeh @vitalune7]

@AnthropicAI can you guys make a night-friendly version of the page first, jesus christ my eyes were burning

[Amir Valizadeh @vitalune7]

@ClementDelangue of course it is. its literally the best the public has ever had access to. i started pulling weights as soon as i noticed the launch lol
https://t.co/sRwdHytGHg

[Amir Valizadeh @vitalune7]

kimi k3's open weights are live on hugging face.

who knows how long they stay there once the closed labs start making calls.

don't guess. mirror them somewhere rn with enough disk.

this is the most capable intelligence humanity has ever been able to download, access, and own.

[Amir Valizadeh @vitalune7]

@sudoingX immediately started pulling weights as soon as i saw the release

am so grateful to have 5tb google cloud storage as a student
https://t.co/sRwdHytGHg

[Amir Valizadeh @vitalune7]

kimi k3's open weights are live on hugging face.

who knows how long they stay there once the closed labs start making calls.

don't guess. mirror them somewhere rn with enough disk.

this is the most capable intelligence humanity has ever been able to download, access, and own.

[Amir Valizadeh @vitalune7]

kimi k3's open weights are live on hugging face.

who knows how long they stay there once the closed labs start making calls.

don't guess. mirror them somewhere rn with enough disk.

this is the most capable intelligence humanity has ever been able to download, access, and own.