Viktor

Detective 🕵🏻‍♀️🕵🏻‍♀️🕵🏻‍♀️

@brxnedotmov Agree. Absolutely amazing brain exercise! -loved every part of it.

@slopshipper Yeah but the general idea stands, and that's very important.

Fable lie to you as much as his maker dario sucking souls amodei -and then fucking anthropic people gonna ask you “why are you hating on them” - meanwhile serving you 10 reasons why for breakfast every single day for over a year - you have low self respect if you ever touched these spyware models that are purposely made in such a way to poison your brain into a fucking ai psychosis until they have you hooked on like a fucking crack addict

Better coordination layer could make existing talent far more effective. Idisagree with the idea that talent automatically flows where it benefits everyone. Talent usually follows incentives, compute access, reputation, available time and legal freedom. Many researchers also cannot freely contribute employer-adjacent work because of IP andconflict of interest constraints. So the platform would need more than git and a package manager. It would need strong incentives, attribution, reproducibility standards, governance and access to compute. If those parts are solved, then yes, it could meaningfully reduce wasted talent.

The bottleneck is basically unorganized P2P research and documentation. Talent automatically flows where they can benefit everyone else. Working at anthropic at day and opening push requests at night is really something programmers would do for fun. Just like linux. Altho whole different approach.

@brxnedotmov 😎😎😎

@slopshipper That's good - engagement is necessary for progress.

You can clearly see when im waiting on the machine to do the task by my spam bursts here 💅💅💅

@brxnedotmov Yeah i cant wait to go through it. Very spectacular. Thanks for sharing this!

@slopshipper The closest we ever been to understanding the universe in its raw form. Very exciting stuff.

Silicon Valley irl.

That clarification makes some sense. An n8nstyle modular layer for ai components, backed by git and a package manager, could be useful infrastructure. Important: ->that is different from saying expertise is not the bottleneck. Infrastructure can reduce duplicated work and make strong modules easier to reuse, but the difficult part is still producing modules that are actually correct, compatible, reproducible and meaningfully better than existing approaches. “Anyone can write a benchmark” is true in the literal sense, but not everyone can design a valid evaluation. ->The hard part is choosing proper baselines, datasets, metrics, controls and interpreting the results without fooling yourself. So i think the strongest version of your argument is: ->open modular infrastructure can increase the leverage of scarce talent. That is a good point. ->but it does not replace the need for scarce talent

Anyone can make a nieche network architecture, anyone can write a benchmark evaluations - the problem isn't lack of expertise, it's the lack of open source infrastructure that can actually reshape the NN space. Lets say a lot of people work on the similar thing. Basically N8N but for AI modules. Imagine you want to integrate context size optimizations - there are multiple different versions and implementations. Same for any other module - basically N8N that runs on git. And has package manager.

@brxnedotmov Oh wow thats cool as hell is!

@slopshipper x.com/i/status/20757… Speaking of

-That could help people discover and coordinate talent, but it does not solve the talent bottleneck itself. A network can connect capable people; it cannot manufacture expertise, judgment or research ability. -You also shifted from proposing a new AI architecture to describing a community and package distribution platform. Those are separate problems. The original technical questions still remain: what is the architecture, how is it trained, how is it evaluated, and what evidence suggests it outperforms existing systems?

A team that consists of mathematics, (physics), med and comp sci students that shares the same passion. AI. This is how everything we loved started : few guys in the college dorm (or something similar)- Facebook, Amazon, Google, Paypal. This is where DAO spaces were meant to be taken. Not to create your own RP server - but to network. It's like building a highway for great minds to connect. Linkedin feels too official. This would be more of a community. Package manager like. After enough time you curate updates. Just like videogames.

@ufc @TheoVon He paid for the full ticket and he gonna use it to the fullest 😎😎😎

You know he's here early 🤩 @TheoVon | #UFC329

I also I think this underestimates the human bottleneck. Better architectures, local models and autonomous tooling can increase leverage, but they do not remove the need for scarce talent. Someone still has to define the objective, design the system, evaluate outputs, identify failure modes and make quality decisions under uncertainty. In practice, talent, judgment and domain expertise are often the limiting resources not merely model size or architecture. tools can amplify strong people, but they can also amplify bad assumptions and produce more sophisticated noise So the key question is not only “what replaces transformers?” but also “who is capable of building, directing and validating these systems well?” That human layer is central to everything you described

Possibly. But we need more improvements to AI architectures. Completely forget transformers. We literally walk the vector space with few tricks. Very small but very precisely designed local models. Something that can actually think on its own - no prompts and improving. Prompt - automatic response. Fast, reliable and actually being intelligent. AI plugins are projected to appear in 2027 - which let you as community member to maintain it - or without it through pattern recognition. It can flag AI slop and find real pages - learns based on context. Very exciting things. That would break every AI monopoly - and give people the power to run local models that are better. Who wouldn't be in that market...

@DovySimuMMA As a backup fighter right?

Michael Chandler will be in attendance for Conor McGregor’s fight tonight 😂

@alightinastorm @finkd That motherfucker. 🤬🤬🤬

hey @finkd if you try to be a vibe coder with sunglasses we demand you to share some vibe coded app nobody gets the sunglasses for free brother

Well first thing first is that it’s a fact that anthropic does most evil things in the world. Now technical things ->Small local models, vector-space retrieval, autonomous agents, pattern recognition, and plugin systems are all real areas of development, but none of them automatically imply replacing transformers. “Walking the vector space” is too vague to evaluate technically, and a system that acts without an explicit prompt still needs some combination of state, triggers, goals, memory, feedback, or environmental input. ->The main questions would be: What architecture replaces transformers? How is it trained? How does it maintain context and update behavior? What makes it autonomous rather than simply event-driven?

Just a rat stuck inside @threejs pipelines 🐀🐀🐀

@VoidStateKate @elonmusk @sama He hit him with a “homeboy”, its so over queen 🫠🫠🫠

@elonmusk @sama @sama you better hold up my brother in Christ or else Elon is not going to let you into Mars 😭

homeboy you're the one sellling public market investors on short-term space datacenters

@elonmusk @HannahParkl28 @sama Just kiss each other already guys

@sama We start flying them next year. Maybe you can come see them if your parole officer approves. After stealing an open source AI charity, you then stole all of Apple’s phone technology! Wow. What do you plan for an encore? That’s tough to beat.

Rumours are that tonight after the main event dana white will announce sam altman vs elon musk as a main event of ufc 350

@VoidStateKate @ChatGPTapp Agi.

it's settled @ChatGPTapp ty

@sandislonjsak Np cant wait to see what you build dude!

@slopshipper Nice, thanks for the inputs!

I have about a month to wrap up some projects and I will start building apps in public as a side hustle because I miss coding and building apps. Running an agency is cool and all, but there’s an engineering side of me that can’t be suppressed. I’m actually thinking about building games because I’ve never done that in my life. Do you guys play mini games on your phone to kill time? Is that still a thing?

Yeah then just rawdog it straight to threejs stack. You have JavaScript knowledge -you played games and know how things functions And there is such a cool part of optimising things and creating custom pipelines for the game to run in a browser, basically the best part of engineering Its so cool dude

@slopshipper Especially since I did a shitton of JavaScript anyways