liyu zerihun

Creating things && Undergrad @ UCLA

[liyu zerihun @liyu_zer]

The road to hell is often paved with good intentions.

[Kelvin Roman @KelvRoman]

I named my company Daraja. Swahili for bridge.

For months, that name was an aspiration. Each Babel model spoke one language and one language well. Creole. Swahili. Yoruba. Twelve languages, each a specialist, each connecting only to English.

But Africa doesn't run on English.

A trader in Kinshasa speaks Lingala. Her supplier in Lagos speaks Yoruba. A nurse in Addis speaks Amharic. Her patient's family speaks Somali. The real connections that Africa needs aren't language-to-English. They're language-to-language.

So I built the Bridge - 23 languages supported, any language to any language across Africa, Europe, Middle East and Asia. Live today.

A Wolof speaker in Dakar can now message a Zulu speaker in Durban. Without English. Without a translator. Without anyone in between.

The continent with the most borders now has the fewest barriers.

Live on translate and API

https://t.co/SjA7P1ewVM

#AIforAfrica

[liyu zerihun @liyu_zer]

@KelvRoman Very cool!

[liyu zerihun @liyu_zer]

Time Without Timesteps, a paper on simulating coupled dynamical systems, will be released soon.

It will be a very basic introduction, hopefully, some will find it interesting!

[liyu zerihun @liyu_zer]

@jimstewartson What about training, though?

[liyu zerihun @liyu_zer]

The codex app is great!

[liyu zerihun @liyu_zer]

@OpenAI guys, we need to figure out how, somewhere down the line, GPT became corporatized in its language. It is really bad; hopefully, you can fix it.

[liyu zerihun @liyu_zer]

@aakashgupta Fiscally, the debt is very manageable, and the assets of the country are owned by Ethiopians.

In fact, in terms of foreign ownership, Ethiopia is in a much better standing than India.

Your post comes off as shallow and poorly researched.

[liyu zerihun @liyu_zer]

@aakashgupta Chinese debt in Ethiopia accounts for roughly 1/4 of the total debt, which is substantial, but not as high as you made it out to be.

The picture gets even more nuanced when you look at the split.

The reason why Ethiopia defaulted was due to a civil war.

[liyu zerihun @liyu_zer]

@HiSohan No problem, good luck with the project!

[liyu zerihun @liyu_zer]

@HiSohan The reason I say that is because I know a bit about how the simulations run, and Jaxley immediately came to mind.

Not trying to downplay your work at all. I mean, I would love to see it and play around once it is done.

[liyu zerihun @liyu_zer]

@HiSohan I understand. I have also run neuron simulations, but regardless, you cannot say the field was bound to running on a CPU, given that prior work already exists.

Yeah, it is a post, but the details matter.

[liyu zerihun @liyu_zer]

@HiSohan Jaxley is also in Jax, in the name.

You did not mention Jaxley once; there are already frameworks that provide GPU-based differentiable simulations of biological neurons.

It seems what differentiates this is speed, which is very good, but you have to be clear about that.

[liyu zerihun @liyu_zer]

The biggest bottleneck for this approach is compute availability.

Which should tell you something.

[liyu zerihun @liyu_zer]

Today I realized,

The most promising alternative to the current trajectory in deep learning is large-scale brain emulation.

Imagine you could replicate a human brain into a machine, augment it, train it, and run it quickly.

This future is not too far away.

[liyu zerihun @liyu_zer]

Sometimes we forget that some of our theories are destined to be very, very wrong.