2025 introspections: my year back at Google
It’s been about a year since I returned to Google, and I thought I’d share how the year went from a personal POV. It’s also the end of the year so I feel somewhat introspective.
This was an interesting year with many notable events, including a couple of career highlights I’m quite happy about. So here’s a “low key” blog post about how things are going.
While most people know me in the past for my work on pretraining and architectures (along with scaling and objectives), I’ve mostly transitioned to working on RL and reasoning. I’m still a “modeling” guy and I think nothing much as changed about that, and RL is basically the technical canvas which we innovate upon these days.
One major technical highlight for me this year was probably co-captaining (i.e., co-leading model training) for the Gemini model that achieved the IMO gold medal (that shortly launched broadly as the first Gemini 2.5 Deep Think model). Being physically in London during the culmination of the IMO project was also a very interesting experience. If you lurk around my socials, you might find the lore and fun adventures behind this model.
Looking back, I was really glad to have been given an opportunity to be part of this historic win. Moreover, if you asked someone 5-10 years ago what getting to AGI would mean, they might have said a gold medal at the IMO would be it. I'm not here to debate the definition of AGI, but this was definitely a very significant and historic moment for AI.
Overall, I was genuinely happy and at peace to be back at research, back to TPUs and back to this legendary infrastructure. I was also very glad to reunite with my past favorite collaborators and good friends. A lot has happened this year, I was proud that some of the new bold research that we had tons of fun with shipped in the Gemini models at Google I/O; along with making a pioneering contribution to some fundamental architecture change to Gemini3, amongst many things which I cannot say too much about publicly.
I have to say the AI vibe is on-point these days. AI is also getting really good. I’ve been using AI more and more in my daily life - Antigravity is really good and I use it on a daily basis. Nano Banana was probably the best AI product/model that came out this year. I think we’re in a cool inflection point where AI is starting to get really really good. Really looking forward to what 2026 brings!
—
Looking back at this year, I’m really glad I spent the whole year as an IC; heads-down on technical work. I think the IC lifestyle is enjoyable and probably optimal for happiness. It is a blessing indeed to be able to probe at the mysteries of the universe with your very own hands and with code as the medium.
That said, with the natural progression of time, most of my friends are now managers with teams of their own. So it’s something I consider every now and then. On that note, most people know I'm based in Singapore. Even though I don’t identify with the local research community (I work nights with folks from Mountain View, NYC and sometimes London), starting a super strong and AGI-relevant team here was something discussed often between Quoc and myself, even before I returned to GDM.
People are interested in a cracked team in Singapore working on the true frontiers of AI because nothing close to this exists in the region. There is also significant untapped talent that requires the right research taste and technical competency to tap into, which is great for GDM too.
I looked around and immediately knew I had to be the one to build it, even if it meant sacrificing a tiny bit of personal happiness. I have to say though, pure IC life is really amazing. Mostafa also told me that having reports and mentoring is very rewarding, so I’m highly convinced for the time being. That said, I’ll think I’ll still want to be mostly focus on pushing things with my own experiments and hands.
The good news is that the gears are starting to spin, and I think we’ll have an outrageously strong team soon and the first ever frontier lab in Singapore. For those curious, the launch of this new lab/team can be found here (x post, linkedin post).
Related to this, a big fun highlight this year was organizing the Gemini Singapore summit (x post here), where we hosted Jeff Dean, Quoc Le, Benoit Schillings, and Denny Zhou in Singapore. Many people told me this was symbolically the defining moment for AI in Singapore, and I wholeheartedly agree.
Oh, aside from inspiring and bringing the AGI aura to this tiny island, and we also got to (privately) meet and chat with the legend Lee Hsien Loong at the Istana. It was an interesting but intense conversation which I think I will remember for a long time. LHL is every Singaporean family’s hero and my parents were surely (kinda) proud of me and honestly kind of shocked and confused when I told them we were going to have a meeting with the legend himself. For some reason, some people still ask me this every now. Was a notable thing this year for sure.
—
This was an overall amazing year. Of course, it could always be better, but it was epic enough for me. What I hope for myself, and for the future team that will fight alongside me, is that we enjoy doing good work and be legendary.
To conclude, I think we’re very lucky to be in the middle of all this: the pivotal moment for humanity and to be actively involved in building AGI. To be able to train models at the literal frontier, to be able to advance the fate of humanity by our ideas and sheer creativity. As AI engineers & researchers at the forefront of this technology, we’re already blessed and already “struck the lottery” (for some, literally a financial lottery lol).
Honestly, I cannot imagine how I got to this point, and if I reinitialize my life I am pretty sure I will not be so lucky as I am right now.
Extended thoughts
One piece of advice I’ve taken to heart recently is that all that matters is doing good work. If you ignore all the noise and annoyances (in life, in work, and especially in AI where the signal to noise ratio is generally low and annoyances are aplenty) and just do good technical work, you will be well respected.
And in order to do good work, one has to be happy and healthy - and therefore keeping healthy is part of the job. I also learned a lot about myself this year: how to “denoise,” how to have more “mindfulness,” and gratitude. Being internally happy is a skill that needs to be learned especially with all the chaos and noise in the AI world. (ZenAI not GenAI haha!).
On health, I’ve also gotten a lot healthier and my healthmaxxing journey has been progressing quite well. In my previous blogpost just a year ago, I wrote about gaining ~15 kgs during the startup journey but I forgot to mention that I’ve been healthmaxxing and exercising very regularly these days.
I am proud to say I have lost almost 25 kgs from my peak weight (84kg -> ~60kg). My resting heart rate (lower is better) has decreased by almost 30 BPM (down from 90 -> 60). My HRV (heart rate variability, higher is better) has gone up by almost 2x. I even took part in—and won—an internal (Google) mixed doubles badminton competition, LOL! Aside from my days in the military, I have to say I might be in my peak health these days.
It’s interesting to take your body as a canvas for trying things out (ablation studies on your own body!). I can kind of get the appeal of Bryan Johnson how he tries out some healthmaxxing protocols on himself. It’s surely interesting. At some point I also took inspiration but how Karpathy writes about biohacking himself which was pretty cool.
—
In the earlier days of my career, and even as a graduate student, I would focus almost all my energy on work. I was an all-or-nothing type of person and maybe thats why I managed to reach this point of where I am today. But i’ve learned, and especially this year, that balance is critical. I don’t mean I have evolved into a weak WLB loving person, but realizing that moving about efficiently matters (compute efficiency lol). There are compounding effects of good health and having a strong support system around (family, friends). As we move into 2026, I want to learn to be more efficiently intense.
