Emir

ai research @uiuc | prev @unibogazici | gen-ai, real-time voice-ai | life is beautiful

[Matthias Niessner @MattNiessner]

The hot topic at #ICCV2025 was World Models.

They come in different flavors — (interactive) video models, neural simulators, reconstruction models, etc. — but the overarching goal is clear: Generative AI that predict and simulate how the real world works.

[Emir @emirkisa]

@joherkhan A social media with strict organic content policy would be fire

[Emir @emirkisa]

I can't stop seeing everything as a distribution after studying diffusion models

[Emir @emirkisa]

@whosetafa :D

[Emir @emirkisa]

“Treat life like a sandbox.
Play it.”

Being young. I love it.

Alchemy - Chris Bartels

[Emir @emirkisa]

@uznr_berkay Very well curated list👌🏻👌🏻

[Emir @emirkisa]

Traditional AI maps a point in the input space to a single point in the output space, essentially producing deterministic outputs for given inputs.

Generative AI maps a point in the input space to a distribution over the output space. This probabilistic nature is what allows generative models to sample diverse and plausible outputs.

[Emir @emirkisa]

A blog post proposing that LLMs are exactly four breakthrougs, each created to unlock a new data source.

1- DNNs (AlexNet, 2012) → unlocked ImageNet
2- Transformers (2017) → unlocked The Web
3- RLHF (2022) → unlocked Human feedback
4- Reasoning (O1, 2024) → unlocked Verifiers

For the vision models, unlocking YouTube *The Visual Web*, could enable the emergence of generalized large vision models. Much like the Web enabled LLMs.

[Jack Morris @jxmnop]

new blog post

"There Are No New Ideas In AI.... Only New Datasets"

in which i summarize LLMs in exactly four breakthroughs and explain why it was really *data* all along that mattered... not algorithms

[Emir @emirkisa]

“The switch to the professionals is a huge step, one that takes a lot of adjustment and a lot of learning.” - C. S. Dweck

[Emir @emirkisa]

A very well curated list of must-read people shaping AI

[Kate Deyneka @katedeyneka]

The fastest way to get smarter is to read smarter people.

I went through @tbpn’s Metis List and pulled together blogs from some of the most influential people in AI. I’ve also added a few gems outside the list 🧵

I find the most value in blogs where people share applied insights and digest SOTA research. Share your favorite AI blogs in the comments!

[Emir @emirkisa]

@katedeyneka @tbpn Great curation 🙏🏻👌🏻

[erkamkavak @KavakErkam]

How can we build a realtime avatar generation model?
I have been exploring this for a while and https://t.co/4Xf2AuOXcv repo is the closest one for this task among my research. Therefore I will try to explain its architecture 👇

[Emir @emirkisa]

The core of modern diffusion generation lies in Denoising Diffusion Probabilistic Models (DDPM) paper.
Understanding it deeply is a cornerstone.

This explainer by ex-OpenAI researcher @lilianweng covers both the rigor and the intuition:
https://t.co/mAMdp8Dm0E

If it still feels heavy, start with VAEs first, she’s got great explainers on those too.

[Emir @emirkisa]

You don’t master something by counting hours.

You master it by running experiments and fixing your mistakes.

[Emir @emirkisa]

@ahmeterdempmk @khanacademy Very well put Ahmet👌🏻👌🏻

[Emir @emirkisa]

@aditiitwt Use opal

[Emir @emirkisa]

Attachment to identity is how people stay stuck.

Don’t confuse your cage with your character.

[Emir @emirkisa]

A few days ago, I had a sweet realization while running in a forest I visit often:

I was thinking about “change” in general while running my usual path. Near the end, I randomly increased my speed beyond my usual pace.

That single change in my internal state “speed”, completely altered my path. Because moving faster meant I couldn’t take my usual sharp corner, thus a new set of possible decisions emerged.

It felt like I was discovering a new forest while still in the old one.

The lesson was: “Even if the environment stays the same, the experience you have can change drastically if you change.”