The Chinese and artificial intelligence

Wolfgang von Goethe

The media has been reporting on a recent breakthrough in LLMs (large language models), the hottest applications in the field of artificial intelligence, which in the last two years have given rise to great hype, suspicions of stagnation and accusations of excessive energy and water costs.

This latest breakthrough has been achieved by a Chinese company, DeepSeek, which has jumped to the front line because it successfully competes with the big companies in the field, OpenAI (which built ChatGPT and GPT4) and Google (with GEMINI), but at a lower cost.

There are now many tools of this type, but the two mentioned in the previous paragraph (GEMINI and GPT4) provided the best performance, although they were the most expensive. Their code is secret, owned by the two companies (Google and OpenAI) and users cannot modify it. Alongside them, now there are others with open source, which can be adapted to each user, but with lesser performance.

It should be noted that the DeepSeek model is only revolutionary because its software is faster at accumulating the data it needs to answer questions, data that (as in all LLMs) are taken from the Internet, and because the hardware uses cheaper second level chips instead of the most advanced. This led to a drop in the share price of NVIDIA, the company that produces the most expensive chips, which may now prove unnecessary.

In an article published on arXiv in April 2024, the team that built DeepSeek explains how they did it and what results they obtained. The Chinese tool has placed itself far above all existing open-source tools, slightly above GPT4, and slightly below GEMINI, as indicated in the attached figure.

I’m surprised that the existence of DeepSeek has not been made public until now, as it reached those efficiency levels in January 2024, and the arXiv article was published almost one year ago.

To test this new tool and compare it to previous ones, I decided to ask the same question, which I mentioned in another post: Who is the author of the following quote: They bark, therefore we ride! When I made the question in English, this was the answer: The quote "They bark, therefore we ride!" is attributed to Genghis Khan, the founder and first Great Khan of the Mongol Empire. It reflects the Mongol attitude toward their enemies and their relentless, fearless approach to conquest and warfare. The phrase symbolizes their determination to act decisively in the face of opposition.

Then, when I made the question in Spanish, the answer was this (my translation into English): The quote “They bark, therefore we ride!” is popularly attributed to Johann Wolfgang von Goethe, although there is no direct evidence that he wrote it. The phrase is associated with the idea of ​​moving forward despite criticism or obstacles. However, its exact origin is uncertain and it could be an adaptation of a similar thought expressed in his work or in the literary tradition.

The English answer is absurd. The Spanish answer is not entirely correct, since there is direct evidence that Goethe wrote the poem Kläffer (Barkers). At least, it is better than those ChatGPT and GEMINI gave me a year ago. For comparison, I asked GEMINI the question again, and its answer no longer attributes the phrase to Trapiello, but (erroneously) to Cervantes, although Don Quixote is not mentioned. And when I ask, Doesn’t it have to do with Goethe? it refers to the poem Kläffer, but to deny that it’s the origin of the saying, and translates the first part of the poem (quite badly, by the way, it’s obvious that it does not translate well from German), but omits the second, which is the part of the poem that served as the origin for the Spanish saying.

This is Goethe's poem:

Wir reiten in die Kreuz und Quer

Nach Freuden und Geschäften;

Doch immer kläfft es hinterher

Und bellt aus allen Kräften.

So will der Spitz aus unserm Stall

Uns immerfort begleiten,

Und seines Bellens lauter Schall

Beweist nur, daß wir reiten.

This is a good translation of the second part, whose relationship with the Spanish proverb is evident:

The dogs would like to come with us

The echo of their barking shows that we are riding

In short, I am still very far from considering acceptable the answers given by these tools, which is logical, given the way in which their responses are generated, which I described in another post: by applying an algorithm that has no relation to truth.

 The same post in Spanish

Thematic Thread about Natural and Artificial Intelligence: Previous Next

Manuel Alfonseca