Trevor Chow
I think about AGI - if you do too, I’d love to meet you!
Previously, I researched inference time scaling at Hazy Research and traded index volatility at Optiver.
I did half of my undergrad at Stanford and half at Cambridge. When I was younger, I was a King’s Scholar and competed in the UK maths and debating circuits.
Outside work, I split my time between ski slopes, skydiving dropzones and the Eras Tour.
Feeling the AGI
Strawberries and the Reasoning Models
o1 is a bigger deal than ChatGPT, because it marks the start of the reasoning paradigm.
Three Revolutions in Pre-Training
The three key insights from pre-training LLMs in the last 4 years is the importance of data.
Inference-Time LLM Routing with Smoothie
It is possible to select the best LLM output at inference time without labelled data. Accepted to NeurIPS ‘24. Blog post here. Code here.
Incidental Causes of Polysemantic Representations
Polysemantic neurons, which are an obstacle to interpreting AI, can arise incidentally. Accepted to BGPT & Re-Align @ ICLR ‘24. Blog post here. Code here.
Transformative AI and Real Interest Rates
The arrival of AGI means an increase in real interest rates. Accepted to Oxford GPR ‘23 and covered by Vox, The Economist, The FT, MR etc. Blog post here.

Strawberries and the Reasoning Models
o1 is a bigger deal than ChatGPT, because it marks the start of the reasoning paradigm.
Three Revolutions in Pre-Training
The three key insights from pre-training LLMs in the last 4 years is the importance of data.
Inference-Time LLM Routing with Smoothie
It is possible to select the best LLM output at inference time without labelled data. Accepted to NeurIPS ‘24. Blog post here. Code here.
Incidental Causes of Polysemantic Representations
Polysemantic neurons, which are an obstacle to interpreting AI, can arise incidentally. Accepted to BGPT & Re-Align @ ICLR ‘24. Blog post here. Code here.