Breaking was passion. Creating became my obsession.

I was a well-known grey-hat hacker who took an interest in the sub-field of artificial intelligence and became obsessed with breaking SOTA.

Understanding how systems work and reverse engineering was the foundation.

I consider myself a lifelong hacker. Before, it was hacking systems and reverse engineering workflows and algorithms. Today, it is breaking down human understanding and hacking it into machines.

I grew up around computers and had access to them from childhood, as my family comes from a background combining software engineering, mathematics and drama school.

I still very much indulge in hacking, particularly DEF CONs and CTFs, while keeping my obsession with the sub-field of artificial intelligence as my primary focus.

I came to Artificial Intelligence from absolute tragedy.

I wanted to become a neurologist, but a high-school crush on an Iraqi girl made me reconsider my prospects in Computer Science. She wanted me to study something more presentable than hacking and become a cybersecurity researcher.

The rest happened because I saw the respect a top researcher receives at places like ICLR, ICML and NeurIPS. It instantly told my young mind: I have to become one of them.

Initial experiment.

By seventeen, I was working at the Harvard University-funded Harvard GAMI lab. I worked on projects including cell segmentation in data-constrained settings with Prof. Anna Yaroslavsky, bone fracture detection, and a reward-hacking chatbot with Dr. Kiran Agarwal-Harding.

I was leading a team that included Luca Leibi, Emma Farina and a few others, among whom some are now doing their PhDs at Harvard University.

Intelligence was bigger than AI.

I went to take an internship at the Genzel Lab because I had missed the window to apply for the Janelia Research Campus by HHMI. I applied to work with Prof. Lisa Genzel and, after a few months, resigned.

I then continued at the Donders Institute for Brain, Cognition and Behaviour as a guest researcher with Prof. Paul Tiesinga. I worked with him for a few months before leaving.

Discovering absurd scale.

Leaving Donders, I found LAION through collaborating with them on LAION-Debate. That led me to work with Jenia Jitsev on EuroHPC supercomputers and EU grants, while being supervised by Christoph Schuhmann across different LAION projects and building internet-scale engineering infrastructure.

I was a key part of multiple projects, including Project Alexandria, LAION-Debate, the LAION-Intel collaboration, internet-scale download projects, LAION-Tunes and more.

LAION ↗

Departing LAION and ELLIS Influence.

I left LAION and ELLIS in quite an unfriendly manner. We are still friends, however, and I met many wonderful people there. Several of them now work at Cloudflare, Google DeepMind and OpenAI, while others lead their own teams as postdocs.

At LAION, a few of us had started Sleeping AI as a joke, where we would publish experiments. When I left LAION, Sleeping AI transitioned into its own independent entity.

Sleeping AI has a mixed reputation regarding the research sub-fields it chooses, but it focuses on producing technically strong work and SOTA contributions.

Small teams. Large problems.

Sleeping AI rethinks the LAION mentality, but we lean heavily on high-value technical collaboration because we deliberately want to keep teams small and produce results faster.

We currently work with Kandinsky Lab in Russia on multimodality and audio models, evaluations and datasets. I moved towards collaborating with people fanatic enough to believe that we can answer P versus NP.

Achieving SOTA
isn't the objective. Finding answers is.

Contact ↗