A recent introduction to Microsoft's BitNet paper reminded me of a segment in the history of computer development.

Microsoft published papers on "BitNet" consecutively in 2023 and 2024. In these papers, Microsoft proposed a new method for training large language models (LLMs). This new method eliminated the vast amounts of expensive multiplication operations that are currently necessary for training large language models, replacing them with simple addition and subtraction operations. This method drastically reduces energy consumption and the huge demands on storage.

Experimental tests showed that this method could increase the running speed by 2-4 times based on the current standards, and it can even run large language models on ordinary CPUs.

The key to achieving this method lies in replacing the current binary numbers (0, 1) with ternary numbers (-1, 0, 1).

After reading this introduction, a great invention historically referred to as the "lost ancestor" immediately flashed in my mind.

This invention is the world's first ternary computer, Setun, invented by scientist Nikolay Brusentsov from Moscow State University in 1958.

Coincidentally, the ternary system used in Setun is exactly (-1, 0, 1), the same as the base structure of BitNet—this makes one wonder whether the design of BitNet likely referenced Setun.

That was during the Cold War, when the rivalry between the US and the Soviet Union had widely impacted all aspects of politics, economy, and science. However, political opposition could not impede global scientists' immense curiosity about technology.

The unusual thought process of Soviet scientists still garnered significant attention from the Western tech community.

After this computer was introduced, Western countries immediately purchased 50 units for experimentation.

Domestically in the Soviet Union, the practical performance of this computer was also quite remarkable.

At that time, the reputation of Soviet light industrial products was usually that they were expensive and of poor quality.

However, this computer exhibited excellent quality, especially displaying extremely high reliability. While other Soviet binary computers of the same era could only operate continuously for about 60% of the time, this computer boasted continuous operational time of up to 95% - 98%.

Unfortunately, such a great invention that could have changed the trajectory of digital age development perished under the Soviet bureaucratic system for two main reasons:

- Low price leading to "no profit"

In the planned economy of the Soviet Union, enterprises preferred products with high prices and large amounts when conducting efficiency calculations. This is because such products could better reflect the achievements of officials in numerical form when calculating GDP. Thus, Soviet officials and factories generally favored "stupid," "large," and "rough" products.

Setun had an extremely low cost, requiring only 27,500 transistors at the time.

How could such a low-cost product reflect the achievements of officials? Therefore, officials were rather dismissive of it.

- Political barriers

Setun was entirely a serendipitous invention by this scientist from Moscow State University. It was not in the university's research plan at all. Not being in the plan meant it posed significant trouble for the Soviet Union's Ministry of Radio Technology. The leaders in the bureau were at a loss as to how to handle such an unexpected invention to meet the higher-ups’ expectations, so the simplest solution was to completely ban the project.

- Influx of orders

In fact, the orders coming from the West far exceeded 50 units. However, economic interests were not the primary consideration; thus, after exporting 50 Setun units, the Ministry of Radio Technology rejected all subsequent orders and forcibly banned the project in 1965.

A product that clearly had market competition due to its low price and excellent quality became the "ugly duckling" that nobody loved in a planned economy and was forcibly suppressed.

Returning to Microsoft's BitNet paper.

To continue simulating and training the ternary large model, Microsoft's current approach is to develop a simulator on existing computing systems, allowing our general binary computers to run a ternary simulator, and then run BitNet within this ternary simulator.

I discussed these issues with Gemini, and its response was about what I expected:

The ternary computer is the ideal platform for running the BitNet system.

Of course, these reflections are purely my personal insights.

History is history.

But looking back at this historical segment, I understand even more the fundamental viewpoint expressed by Mr. Buffett when discussing his ultimate confidence in the US stock market and American companies:

As long as the US continues to uphold the rule of law and a market economy, he believes that the US stock market and American companies will certainly have a bright future.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。