Share this post

SFT: Sonia Fine-tuning
SFT: Sonia Fine-tuning
Farewell OpenAI

Farewell OpenAI

I'm leaving OpenAI Research to tackle the hardest real-world problems in AI performance

Sonia Phene's avatar
Sonia Phene
Jun 27, 2025
35

Share this post

SFT: Sonia Fine-tuning
SFT: Sonia Fine-tuning
Farewell OpenAI
Share

After a wonderful ride on the rocket ship that is OpenAI, I’ve decided it’s time for new adventures.

It has been the privilege of a lifetime to be part of the evolution of ChatGPT and AI during an era that has transformed not just the world, but also me. I learned what it means to do AI research at a time when model capabilities are leaping forward, and watched these breakthroughs continue to transform society.

A few datapoints that capture the scale and momentum:

  • From 0 to 1 billion weekly search queries. I worked on post-training our search model from its earliest launch through rapid growth, contributing to a fundamental shift in how people access information.

  • From 4o to o3. My first model release was GPT-4o for free users; my last was the launch of the o-series reasoning models. Those two launches bookended my time at OpenAI and captured how quickly the frontier was moving.

  • 3× team growth. OpenAI more than tripled in size during my time there, bringing in an extraordinary influx of talent, energy, and ideas.

As a researcher at OpenAI, I had the chance to work on some incredible launches: the Search beta and GA (before full integration), the o-series reasoning models, advanced voice mode with Search, and reinforcement fine-tuning. These projects brought me into close collaboration with brilliant researchers as well as deeply talented engineers, product leaders, and designers. I learned so much from all of them.

The people were by far the best part of my experience. I’ve never worked somewhere with such high talent-density and motivation. I have so many fond memories of late-night Slack pings and walks to Sightglass coffee as we pulled off launch after launch of ambitious, fast-moving work.

Next up, I will be continuing to collaborate with folks I met at OpenAI. I’m eager to dive deeper into the enterprise space and work on bridging the real-world gap between models that perform well on paper and those that succeed in practice - a space full of interesting research problems that haven’t yet been fully solved. I’ll share more on that soon.

But first, I’ve been taking a few weeks to read, meditate, and write. I’ve started this Substack where I’ll be sharing thoughts on AI, along with career advice and lessons I’ve learned along the way. I’d love to have you follow along.

Thanks for reading SFT: Sonia Fine-tuning! Subscribe for free to receive new posts and support my work.

35

Share this post

SFT: Sonia Fine-tuning
SFT: Sonia Fine-tuning
Farewell OpenAI
Share

Ready for more?

© 2025 Sonia Phene
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture

Share