Science is being undermined by the “slow system”, and most people haven’t realized it.

In the past week, several major events occurred in the tech world: GPT-5 was officially released, with its multi-modal capabilities surpassing those of human experts; China’s “Jiuzhang №4” quantum computer broke the computing limit; AI Agent took over a real biological experiment for the first time; the Starship V3 is about to undergo its most critical test flight.

Each breakthrough is telling us that the productivity has shifted gears.

But at the same time, a more hidden and more dangerous rift is widening — the “infrastructure” upon which science operates has remained almost untouched.

Crack 1: Computational power centralization, the academic community is being pushed out of the game

Anthropic has obtained the exclusive access rights to SpaceX’s supercomputer, with 300 megawatts of computing power exclusively for one company. The Dark Side of the Moon completed a $2 billion financing round, with its valuation exceeding 100 billion.

Top computing power is becoming a “private oilfield” for a few players. Academic laboratories, non-profit organizations, independent researchers — they can’t get the H100 cluster, can’t train cutting-edge models, and even can’t replicate those experiments that “only the giants can handle”.

Computing power should not be the ticket to enter the field of research. It should flow freely like water, rather than being monopolized like oil.

Openness is the first principle for scientific research infrastructure.

Crack №2: Lack of trust in the infrastructure, AI becomes a “black box reviewer”

AI can assist in paper evaluation, fund review, and even directly score. But no one dares to use it.

Because we don’t know what the basis of AI is, whether the training data is biased, or whether the conclusions can be verified. Traditional peer review at least has the aspect that “people can be held accountable”, but the results of AI review are opaque and the responsibility cannot be attributed.

This is not a reason to reject AI; rather, it is a warning bell that we must provide a reliable foundation for AI.

Blockchain can provide an unalterable audit trail for every AI judgment: the data source, model version, and reasoning path are all recorded on the blockchain. By involving AI in decision-making, it also enables accountability for the AI.

Crack 3: Speed mismatch, “fast system” is killed by “slow system”

GPT-5 is updated once a week, while research funding comes in twice a year, with each round lasting half a year. AI can read through a thousand papers in just a few seconds, but negative results are never published — the pitfalls others have fallen into, you have to fall into them again.

An absurd situation has emerged:

AI has made “coming up with ideas” hundreds of times faster, but “verifying ideas” is still at the same speed as before.

The more efficient the former is, the more of a bottleneck the latter becomes.

This is not an issue with AI; it’s a problem with the system. The way research is organized can no longer keep up with the pace of AI.

OPENSCI is building the third road.

We do not deny these cracks. On the contrary, we are here precisely for this purpose.

OpenGrants: A decentralized computing power and funding network, enabling good ideas to directly receive support from the global community without the need to wait for the semi-annual review cycle.

IDA + PoSR: A reliable contribution record layer, enabling every piece of data, code, and review comment to become traceable and referable digital assets.

ASCI: Research intelligent agents embedded in a trust infrastructure — not just “smart”, but also “trustworthy”.

A single breakthrough is impressive, but a systematic breakthrough that can be achieved is what truly represents the future.

When AI finishes reading 10,000 papers in one day, how long can the “slow system” of scientific research last?

It won’t last long.

So we must build a new one right now.