In 2026, only language models should extrapolate

I began 2026 with a lot of new things, the way every morning you wake up to a brand-new day. The strangest of them all is that I've gotten one foot inside an old dream: taking my master's eldest son as my own disciple. It's a dream I first sketched out at least ten years ago, in the very moment a friend mentioned that my tech master's eldest son was showing signs of autism, while we were drinking together at some place on Lương Hiền or Hoàng Diệu, I can't remember which. My native instinct is a soul devoted to living beings that need help — the nameless blade of grass that nobody wants to keep, the flower whose smell everyone hates. Because I too have come close to losing my own life a few times, and had it not been for help, intentional or accidental — the first time being before I was even born.

Don't ask how I know I'll be able to do it. I simply want to, then must, then believe that I will. How, we'll figure out as we go. I kept figuring and couldn't find an answer. Honestly, I got stuck, so I just stamped the dream and locked it in the safe. Then one day at year's end, we brothers got together again to talk — springing from the custom of looking back on each year when we used to work together, which had grown into a whole ritual of sitting over tea, chatting about mountains, forests, seas. And then the opening came when he told me about his friend, about the field this friend had chosen to study, about the role he wanted to play on the road called contributing to society — the shortened version being "career." Luck appeared, and Hữu Lộc just had to grab it — luck is Hữu Lộc, my father used to say: whatever you want, someone will bring it to you. So I took it. Although of course I still had to convince the older brother: let him come to me, I've had the welcome mat laid out for ages, and I've even got a new mat now called artificial intelligence.

For now I'll just call him "em" (little brother), but when the time comes I'll have to address him as both nephew and little brother at the same time. My wife says I've reached the age where I get to work with my friends' kids.

His soul is deeply in tune with mine. Combined with years of careful preparation, I'm starting my way of teaching all over again. I got to start teaching the young before I was even thirty — my students back then hadn't yet graduated — so I had a little teaching playbook put aside for safekeeping. The further I went with that playbook, the more thorny problems I ran into. Was this really the limit of my skill as a teacher? Thinking it over, maybe I should try learning again, with him. So I just went for it: in this strange, emotionally turbulent phase of both technological and personal explosion, I have to constantly ask myself what I'm feeling; so I started opening our review sessions with him with the question: how are you feeling right now?

Start by naming the feeling. Then work backwards, telling the stories that led to that feeling. Go all the way back to the starting point of the event or action. Because feelings always travel alongside the journey of thought and action — because every outcome and every value is, at bottom, a feeling.

Like that children's song Nai sings so well (naming your feelings), you need to know what the feeling is, then understand it, then cherish it, and then use it to transform your next action. Positive feelings build resilience for the hardship ahead; negative feelings teach effectiveness, focus, and the awareness that you don't yet know. An endless cycle, the way day and night take turns playing the lead in twenty-four hours.

Only one month in, and I've learned so much. It feels like a baseball player switching from batting to pitching. Just twenty more years of training and I'll be unkillable in the league. I've thanked him two or three times the way you never quite know how much is enough. I've seen my old weakness: I focus on the task, analyze the process, and then I have to extrapolate the feelings that my students are trying to hide. In 2026, the only thing that should extrapolate is language models using computers. For me, I have to know.