What he said
Shou Zi Chew took the stage at a DBS banking conference in Singapore on Monday and did the CEO thing: he predicted the future. Within two to three years, he said, AI will deliver "a real scientific breakthrough: physics, biotech, chemistry, material science." "It's very, very, very possible," he added.
He also offered the standard two-part hedge: short term, we overestimate; long term, we underestimate. The line is older than TikTok. Bill Gates said it in 1996 — overestimate the change in two years, underestimate it in ten. Chew is just the latest executive to carry it into a keynote.
The company behind the quote
Context is worth more than the forecast. Chew was not speaking to scientists. He was speaking to bankers, as the CEO of the company whose app Western teenagers open first. ByteDance's Seed research team did unveil Protenix-v2, a biomolecular structure-prediction model, in April. That is real. But the rest of the pitch — "exploring ways to collaborate with scientists," "some work that we've done in this area" — is the language of a consumer company asking to be read as a research house.
And nothing in the talk grades itself. Physics? Biotech? "Very, very, very possible" covers every outcome and commits to none. Come back in 2029: there is no prediction here precise enough to be wrong.
The half worth keeping
The interesting sentence is not the prediction. It is the warning. Chew said users will hand AI agents broad permissions — including the right to make purchases — without fully grasping what they granted. "The cybersecurity risks are quite big," he said.
That one lands. Last week, Axios reported that executives at OpenAI and Anthropic are privately war-gaming "the day after" a catastrophic AI-driven cyberattack, which industry insiders expect within six to twelve months ([we covered it here](https://alfw.ai/article/openai-anthropic-day-after-cyberattack-drills/)). Chew's breakthrough is due in three years; the labs' disaster rehearsal is on a months-long clock. The contrast is worth more than the forecast.
How to read it
Treat conference predictions as marketing with a science label. What moves is money and published work: the $1.8 billion that Meta, Google and the U.S. government committed last week to a [virtual-cell initiative](https://alfw.ai/article/biohub-virtual-biology-1-8b/), or the models labs actually release. Protenix-v2 was April. If Chew's curve is as steep as he claims, the proof will show up as papers and products, not quotes. Until then, "very, very, very possible" is a forecast that costs nothing to make and nothing to miss.
Sources
- [1] CNBC (syndicated), “TikTok CEO sees major scientific breakthroughs from AI within three years” (October 12, 2026)Read source
- [2] Zotpaper, “TikTok CEO expects AI to deliver major scientific breakthroughs within three years” (October 12, 2026)Read source
- [3] Longbridge, “TikTok CEO sees major scientific breakthroughs from AI within three years”Read source