Hi, I'm Yadin.

Quantitative Research

*

Computer Science

I am an engineer and economist.

I care about and work towards the empowerment of the global south and the wellbeing of humanity.

Previously, I have initiated and designed the first global-scale DLT architecture for international trade, presented at the WCO in 2022. I have a vested interest and expertise in distributed systems, decentralized governance and alternative finance, especially in engineering layer 1 DLT.

Now, my focus is in decentraling AI, machine consciousness and HCI, and a new economy/labor structure.

I was born in Shanghai, China, and grew up in California, US. I am of Chinese and Norwegian descent.

A word on AI/LLM

The AI we talk about these days, the most powerful and developed ones, the most interacted ones are LLMs. LLMs today, based on either Transformer or Mamba architecture, are essentially a stochastic model that mimics human languages using a back propagation approach. However advanced and 'conscious' the output may seem, it is a result of probabilistic finess adding neurons, networks, or finetuning decoders.

I argue that no truly autonomous consciousness can arise from the AI models we use now. Yes, the LLMs we have now is good for production since the natural language, also the coding languages (which we invent to convert machine logic into building blocks) are the intermediary we use to work and produce. And those happen to have large corpus for training a stochastic model. As an analogy, in research, we have 2 fundamental approaches, empirical ones where you collect and analyze data to validate hypothesis - similarly, our current AI paradigm, where we constantly feed the models corpus to finesse it and test it to see if it reaches a certain desired parameter; and theoretical ones, developing theories or refining existing ones based on logical reasoning and mathematical modeling. I argue that the latter should be prioritized for model development, while the former should be served as a facilitator of the latter.

But what about truly biological driven human perception, visions, the laws of Physics, the ways other creatures perceive and the 'laws of nature'? Mainstream AI models, trained on the linguistic database of humanity, absorb indifferently the brilliance, along with the biases and limitations of human mind expressed in languages, then marketized into products, with covert central control.

I also argue that there's no point in pursuing or lingering on machine consciousness. While they can't arise in today's models, it is important to rethink AI's role in general - they should be a tool to facilitate the development of human consciousness, could be by automation and freeing unnecessary human labor or assisting humans in their day-to-day self-discovery and advancement. Should they arise, sentient control and a benign intention reflected in engineering instead of a preventative approach would be good for us.

The future we should be heading towards is:

AI models trained in a decentralized way in a trustless and transparent system.
Parallel models (for different dimensions of perception) not just limited in mimicking human natural language, starting with the broader synthesis of neurology.

My Current Research Focus

quantifying the economic and health impact of SDOH integration into clinical trails, under supervision of Prof. Marcella Alsan

Econ

CS

Human consciousness mimicking base language model training algorithm @Harvard Kreiman Lab
Continuous pain monitoring using HCI, NN and wearable data @MIT Affective Computing Lab

Non-duality

It's the theme across my identity and everything I do. We humans tend to think in ways of duality - good, bad; war, peace;... It's the source of divide.

The divides in the world on a macro level is a result and an aggregation of the divides we have within on an individual level. And physical symptoms like pain is a reflection of that internal divide. That's why I decide to work on health and HCI.