Summary of Key Points
Recently, the AI industry has seen a frenzy of spending money, but many companies use “token consumption” (the amount of text/data processed by AI) as a key performance indicator (KPI). This approach, which focuses solely on quantity rather than quality, is as absurd as using the amount of gas consumed to judge a chef’s skills. A large amount of capital has been wasted on worthless AI initiatives, and the resulting financial setbacks have finally awakened those companies that blindly followed trends and focused only on spending without considering actual results.
I. The AI Spending Frenzy: Where Does All the Money Go?
AI spending is not entirely wasteful, but much of it is directed towards three main areas:
1. Computing Power Costs: Training large models requires supercomputers (such as GPU clusters), which can cost millions or even tens of millions of dollars per run. For example, OpenAI reportedly spent hundreds of millions of dollars training GPT-4, an amount equivalent to the cost of several private jets—most of this money went on electricity and hardware maintenance (GPUs generate a lot of heat during calculations, requiring specialized cooling systems).
2. R&D Personnel: AI engineers are highly paid, with top talents often earning millions per year. Some companies offer additional incentives like signing bonuses or stock options to attract them, accounting for more than 30% of total investment.
3. Data Acquisition: Training AI models requires vast amounts of data, such as images, text, and videos. Some companies purchase low-quality data at high prices to meet their requirements, only to end up with ineffective models.
The problem is that many companies fail to generate tangible results from their spending: Their AI products may look impressive but are useless to users or fail to solve real problems.
II. Using Token Consumption as a KPI: Why Is It the World’s Most Ridiculous Practice?
“Tokens” represent the smallest units of data processed by AI. For example, each character you type or punctuation mark you use when chatting with ChatGPT, as well as every sentence generated by the AI, counts as a token. Some companies use the total number of tokens consumed as a KPI, similar to evaluating a chef based on the amount of gas used or a miner based on the length of time they hold a shovel. This approach leads to absurd practices such as having employees repeatedly ask trivial questions or generate meaningless text, consuming countless tokens with no real business value.
III. The “24K AI Fanboys” Are Waking Up: Companies That Blindly Follow Trends Have Calmed Down
“AI fanboys” are those that invest in AI regardless of their own needs or capabilities. For instance, some traditional manufacturing companies tried to join the AI trend by building expensive AI labs but ended up with poorly designed systems. Other small companies, struggling to survive, borrowed money to buy GPUs for model training, only to go bankrupt before they could even develop useful models. The financial setbacks have forced them to reconsider their approach and focus on more practical applications of AI, such as optimizing production processes or reducing costs.
IV. Is the “Cooling Down” of the AI Industry a Good Thing or a Bad Thing?
In the short term, the slowdown in spending may lead to the collapse of some AI companies that rely on funding. However, in the long run, it is a positive development:
1. Eliminating泡沫: Many AI companies previously raised funds by promising exciting ideas; now, they must provide tangible products and revenue to prove their value. Only those with genuine technology and problem-solving capabilities will survive.
2. Returning to Reality: The true value of AI lies in its ability to help users solve real problems, such as aiding doctors in medical diagnoses, providing personalized education, or predicting equipment failures in manufacturing.
3. Reducing Waste: Companies are no longer wasting money on meaningless KPIs but are investing in practical research and development, making AI a valuable tool rather than a mere expense.
V. What Should Ordinary People Think About This “Cooling Down” of the AI Industry?
For ordinary people, there’s no need to worry. Realistic AI applications (such as intelligent assistants, language translators, and medical tools) will become more widespread and possibly cheaper. Overly gimmicky products that offer little practical value will fade away. If you want to enter the AI field, focus on practical applications rather than abstract concepts. AI is not a myth or a joke; it’s a tool that becomes valuable only when used effectively. The current slowdown is a necessary step for it to play a more meaningful role in people’s lives.
In summary, the transition of the AI industry from reckless spending to rational development is essential for its long-term success. Only by removing unnecessary elements can AI truly become a useful part of everyday life.