@FinanceYF5: Samuel Ting's Four Methodologies 1/ A conference that talked from start to finish about how AI accelerates scientific research, the ninety-year-old Samuel Ting connected from the other end of the screen. What he talked about was not AI, but how a person decides to redo something that everyone thinks has already been done.

Ding Zhaozhong’s Four Methodologies

1/ A conference entirely about how AI accelerates scientific research — and ninety-year-old Ding Zhaozhong joins in from the other side of the screen.

What he talks about is not AI, but how a person decides to redo something that everyone believes has already been completed. https://t.co/bSVKtHnTa5

2/ His criteria for choosing a topic are simple: not “has anyone done this before,” but “if this conclusion turns out to be wrong, what would it shake?”

What he picks are precisely results that the world’s top experts have worked on for years — and that have been independently reproduced.

3/ The second point is even more demanding: you must use an independent method.

Redoing it with the same technique only reproduces the same systematic error.

4/ Third, you must accept the cost in advance.

To run an experiment that others believed already had an answer, he left the United States. To run an experiment that two major laboratories both rejected, he found his own way.

5/ Fourth, the timeline is set by the question, not by patience.

Measuring the size of the electron took eight months. AMS took more than a decade.

The bigger the question, the longer the time.

6/ Sitting in the audience and listening, the most direct feeling was this: in an era where everyone is chasing “faster,” there is still someone seriously discussing “how to confirm that something is true.”

7/ What do you think is most lacking in scientific research today — speed, or this willingness to redo “answers that already exist”?

That’s all.

If you enjoy this topic:

1. Follow me (@FinanceYF5)
2. Like + repost the first post below

a16z on why open-weight models matter: cost and control.

“The world can’t be controlled by proprietary APIs alone.”

“Closed-source models are too expensive, and I want to control the infrastructure and the model myself — whether it’s adding guardrails or scaling it up.”

For a business like voice agents, you must ensure the model responds on time — only a model you control yourself can do that, rather than tying critical infrastructure to proprietary APIs that could go down at any moment.