Stop LLM research? Part -1/2

There are 2 different trends which are alarming even if taken along. But, convergence takes the risks to a completely new level. The 1st issue is speed of innovation coupled with the cost. The 2nd issue is the ego blow to mankind caused by AI. In this article I will address the speed and cost issue.

Historically, innovations which completely changed mankind and civilizations start being very costly and over a period of time became cheaper and more widely adopted. Let me start with a few examples - electricity, automobiles and computers !

Electricity

In the late 1800s, Thomas Edison developed the first practical incandescent light bulb and built the first electric power station in New York City, which allowed electricity to be transmitted over long distances. This led to the widespread adoption of electricity as a source of energy for lighting, heating, and powering machinery.

Initially, the cost of electricity was high and was only affordable for the wealthy. However, as technology improved and economies of scale were achieved, the cost of electricity decreased, making it more accessible to the general population.

Automobiles

We used to have horse drawn carriages. It restricted what was possible - we cant get high raise buildings, because transportation of these materials will be time consuming and expensive. Without, cranes, its impossible to imagine high raise buildings. On top of these it has impacted

1. Urbanization: The rise of automobiles has also contributed to the growth of cities and urbanization, as people are able to travel farther and faster, and cities have been designed around automobile use.
2. Cultural Shift: The automobile has also had a significant impact on cultural values and attitudes, promoting individualism, consumerism, and a desire for personal freedom and mobility.

Computers

• 1950s: The earliest computers were large, expensive machines that were primarily used by government agencies and large corporations. These early computers could cost several million dollars and were typically custom-built for specific applications.
• 1960s and 1970s: The introduction of mainframe computers in the 1960s and minicomputers in the 1970s brought computing power to smaller businesses and academic institutions. These machines still cost hundreds of thousands of dollars, but were much more affordable than their predecessors.
• 1980s: The personal computer revolution began in the 1980s with the introduction of affordable computers like the Apple II and the IBM PC. Prices for these machines ranged from $1,000 to $5,000, depending on the configuration.

In all these 3 examples which completely transformed civilization, the cost was high. It faced a lot of resistance and slowly became cheaper and more widely available.

Contrast this with what is happening now with LLMs. The cost of building these models run into millions. The infrastructure to run these may even run into billions. However, they are made available free (chatGPT, open source LLMs) or provided at throw-away cost.

This creates a dynamic we have never seen before. Transformative technology starting from almost zero cost. This makes the technology uncontrollable - because internet access is widely available and millions of users around the world are going to come up with millions of uses (not all of them positive). It gives society no time to think and adjust.

OpenAI released a document on the impact of LLMs. 80% of jobs will be impacted by these technologies. Just few months ago the US government released a document on AI impact and now that looks so out dated - that is the speed in which AI is changing the job market.

To summarize (mine and not chatGPT) - historically game changing technologies have been very costly and took decades to percolate into mainstream society. In case of LLMs its massively disruptive, fairly unreliable, widely available but almost free. This inverted dynamics in one of the reasons people are calling halt to this technology. In Part 2 I will discuss the ego aspect.

#gpt4 #gpt #chatpgt #llm #capgemini #jobs #openai #ai4good #aiimpact #llms