Giulia Lanzillotta

Full-time dreamer.

AI researcher, PhD fellow at the ETH AI Center. I do continual learning for deep nets.

[Grigory Sapunov @che_shr_cat]

1/Again about continual learning.

What if catastrophic forgetting isn't a bug, but a mathematical necessity?

We build complex systems to retain old knowledge, but new theory proves that in non-stationary streams, perfect memory is an optimization liability. Here is why. 🧵

[Giulia Lanzillotta @glanzillo11]

@che_shr_cat Thank you Grigory! This is a great summary

[Giulia Lanzillotta @glanzillo11]

15/15
I don't have answers. These are simply questions I keep returning to as someone trying to build a career in foundational AI research. I'd genuinely like to hear how others think about them.

[Giulia Lanzillotta @glanzillo11]

1/15
[slow take 🐢] The list of NeurIPS 2026 workshops came out at the beginning of the week (you can find it on OpenReview), and I found it honestly disappointing: so much of our collective energy going to applications and agents, so little to foundational questions.

[Giulia Lanzillotta @glanzillo11]

14/15
I don't blame the workshop chairs: they are responding to what the field is today. But perhaps it's time to recreate these spaces.

[Giulia Lanzillotta @glanzillo11]

@KaiSchmitzHEP @AluLucas1 @fermatslibrary Even the "standard techniques" need to be proved. And these "tricks" are what legitimates their use.

[Giulia Lanzillotta @glanzillo11]

https://t.co/5APWEokCbf
The first article in the deep work series is out

[Giulia Lanzillotta @glanzillo11]

@fermatslibrary I'm sorry I couldn't find any reference for this quote. Where have you found it?