Emad

Building first principles, sovereign AI @ii_posts.

Founder @StabilityAI.

Consistent inference is possible.

[Emad @EMostaque]

The Last Economy is number 1 on Amazon's Best Sellers for AI with a pretty good ratings ✍️

No publisher or big campaign, also available for free on the website and paperback just dropped 📖

Thank you all for the support & the next part is nearly done..🚀

[Intelligent Internet @ii_posts]

Economics was built for scarcity.

The Intelligence Age is about abundance.

Today we release our book, The Last Economy, introducing Intelligent Economics, a unified theory for this new era.

🧵

[Emad @EMostaque]

@saintkamus few months

[Emad @EMostaque]

AMA on Sovereign, open source & frontier AI

[Emad @EMostaque]

@Real__Yahiya @ii_posts We are about to accelerate and release a lot of stuff. Huge breakthroughs & practical on the ground sovreign AI things

[Emad @EMostaque]

@christinetyip @ii_posts It's pretty poor! Update tomorrow on how we will fix it

[Emad @EMostaque]

@CounterplotCo Working on this with state champions

[Emad @EMostaque]

@real_vaishak They will be providers to AI utilities per region. We are building the utility companies at II

[Emad @EMostaque]

@recursive_wave I run II agent ofc plus my own custom harness

I have a multimodel harness that uses a variety of models so don't even really think about them

I use Claude to typeset and tidy latex files and openai to make images/slides though

[Emad @EMostaque]

@sreeramkannan yes! one trick is.. vary the temperature ^_^

[Emad @EMostaque]

@RW_Laws @drjamesgood pretty sure they are of different substance in all ways

[Emad @EMostaque]

AI is heteroousios with Man, as
Man is heteroousios with God

[Emad @EMostaque]

@aaronburnett I just said we found the most passionate and original thinkers which is why we could build sota model after sota model.

Big lab folk could perpetuate but rarely innovate reliably

We ended up with 16 out of 80 dev/researchers being phds

[Emad @EMostaque]

@sreeramkannan ..there's more to improve :D

[Emad @EMostaque]

@sreeramkannan Well smaller models should be more capable of RSI than larger ones right?

Stable diffusion was probably the first one if you consider the ecosystem

Ecosystem trained loras to theri preferences.

Some of those loras were combined to improve the base model => loop

Slow but good

[Emad @EMostaque]

@Real__Yahiya Just use agentic tools a few hours a day. Exercise it like your lats.

[Emad @EMostaque]

@Real__Yahiya Join nemotron alliance. Post train on as much data as you can find

[Emad @EMostaque]

@Zenpai_1 Yes, autoregressive models can only take you so far but tbh thats pretty far!

For competent intelligence they are enough

[Emad @EMostaque]

@JackGeorge17 What would you do if you had unlimited ocmpetent digital and then physical workers

Simple business is just helping people undersand this

[Emad @EMostaque]

@gonlenidefi Open weights models are 6 months from the frontier nad I xpect that to close

[Emad @EMostaque]

@raahilgadhoke Better algorithms on top of better data. Actually using more of the latent space, eg it makes no sense to use the same temperature for everything.

[Emad @EMostaque]

@TheAIShrink Not your models not your mind

[Emad @EMostaque]

@Flandermaxx Tbh nobody has asked about tokens!

Open source tokens we must improve ofc

[Emad @EMostaque]

@PuraVidaFun Yes we have done that and will be releasing soon. Lots of breakthroughs in these specific areas.