The US and China successfully negotiated a tariff deal, to temporarily lower the tariff against each other for 90 days:

• US reduce tariff on China import into US from 145% to 30%; and

• China reduce tariff on US import into China from 125% to 10%.

It is interesting how we sort of “predicted this”, as mentioned in our 9th Community Meetup held on 28 April 2025, where we said that:

However, this is not the end, it is just a half time break. The US had at the same time opened several security investigations on:

1. Imported planes, jet engines and other aerospace parts;

2. Pharmaceuticals including finished generic and non-generic drugs;

3. Semiconductors and semiconductors manufacturing equipments;

4. Medium-Heavy duty trucks and truck parts; and

5. Processed critical minerals

In the announcement of the tariff deal, US Treasury Secretary, Scott Bessent added that the future decoupling will be based on items that are of national security interests:

On the semiconductor front, the US is also looking at overhauling regulations on the export of semiconductors used in AI, which potentially overwriting the AI diffusion rule launched by Joe Biden.

All in all, the progress in the US-China Trade War is moving in the direction to impose product/category level tariff, as expected and mentioned in our previous article “Impact of Tariff & Trade War on Magnificent 7 Companies”.

With 90-days certainty on the tariff war between US and China, and the potential revamp of the AI export regulation to China, the competition in AI might enter the next stage soon. Thus, we want to lay the groundwork on all you need to know about AI, so that you can invest for the future.

Chapter 1: The release of ChatGPT 3.5

On 30 November 2022, ChatGPT 3.5 was publicly released to the world. It takes ChatGPT:

• 5 days to reach 1 million users

• 2 months to reach 100 million monthly active users

• 2 years 5 months to reach 400 million weekly users (as of Apr 2025)

The launch of ChatGPT shocked the world because of how humanlike the AI chatbot is, and also how revolutionary this technology will be to humanity. People were saying that ChatGPT will replace white-collar worker in the near future.

The financial market and Silicon Valley were very excited about LLM and started an AI-craze. The biggest US Tech companies, except Apple started throwing money (capital expenditure) to invest in AI:

All the Big Tech US companies are also rewarded for investing heavily in AI, with share price increasing from 49% to 468% since the launch of ChatGPT 3.5 on 30 November 2022. Coincidently, Apple who invest the least in capital expenditure compared to other companies, has the lowest share price return among all the big US tech companies.

Chapter 2: How Generative AI work

Large Language Model

The technology behind ChatGPT is Large Language Model (LLM). LLM is different from previous AI technology with 2 key breakthroughs:

1. Transformer Architecture that is built to process sequential data like text, grammar, conversation etc by ranking the importance of different words in a sequence, so that it can understand relationship between words. For example, “I am…” is more likely to happen than “I is…”, so the AI will know that it needs to generate “I am…” instead of “I is…”.

2. Probabilistic Manner in Generating Response: The AI will then use this understanding to predict the next word based on probability. When you ask a question in ChatGPT (input), ChatGPT will generate a response one word at a time by calculating the next most probable word, then the next, and so on. Because of the probabilistic manner, the answer can be different every time, even though the question is the same, which makes ChatGPT human-like, and different from normal chatbot which always give the same answer for the same question.

The combination of Transformer Architecture to understand large set of data (input), and the probabilistic manner in generating response (output) makes LLM a better AI system than old AI system that is good at handling large number of data, but can only generate a fixed set of output, which makes it very robot-like.

In short, LLM is an AI system that is very good at:

1. Analyse large number of data

2. Recognise pattern of sequential data (data can be text, chemical, image etc)

3. Generate new set of output based on input

Training

All Generative AI model today, i.e. ChatGPT, Gemini, Claude, Anthropic, Grok etc use LLM as their foundation technology. But, did you notice that there is ChatGPT 3.5, then ChatGPT 4.0, and currently ChatGPT 4.5?

Subscribe to Community Subscription to read the rest of the article!