The AI trends for the year 2025

The year 2025 will bring some significant changes. The reason is the massive further development of AI models and AI systems. What changes will there be and what impact will they have?

Introduction

The following forecasts on AI are based on observations over the past few months. The observations also include an extensive, daily analysis of open source AI models, AI frameworks and AI systems. Considerable technological progress has been and continues to be made on a weekly basis.

The forecasts for 2026 can be found here:

What will happen in 2026?

Companies that do not yet have an AI server should get one soon (NVIDIA consumer grade GPU). Individuals should look for AI-enabled PCs or laptops (NVIDIA GPU or Apple golden cage).

Not much time has passed since the beginning of the AI age. Nevertheless, incredible advances in the capabilities of artificial intelligence have been observed every few weeks or months, especially in the previous year. There is no end in sight (at least in terms of the development of possibilities).

All statements are to be understood as subjective forecasts. The forecasts are given in indicative form and not in subjunctive form.

Forecasts for the development of AI in 2025

Size of language models

Language models (LLMs) are getting bigger and smaller at the same time.

Big Tech companies like OpenAI, Google, Meta or Microsoft will be producing increasingly larger LLMs. ChatGPT 5 has already been announced. However, this won't go on indefinitely. On the one hand, the resource expenditure is enormous. On the other hand, smaller and more powerful LLMs are emerging.

Larger AI models will continue to exist as long as it is technologically possible. The main motivation is achieving an AGI, a superintelligence. ChatGPT has already achieved a score of 87.5% on the difficult ARC-AGI Test, which should be far more than what the average human can accomplish. Note that the ARC-AGI Test what designed to be much easier for humans to pass than for AI models.

Language models and other types of AI models are becoming smaller and smaller at the same time. Current powerful representatives of this type of model already exist. One prominent representative is Llama3.2:3B. The smaller models are becoming increasingly powerful for two reasons:

1. Algorithms and training methods are constantly improving.
2. Large models are used as teachers to siphon knowledge towards the student model: knowledge distillation.

Consequences:

• AI will find its way into every type of end device: not just mobile devices, but even mini-computers such as the Raspberry Pi
• Open source AI will outperform ChatGPT
• AI is getting faster and faster, but not with ChatGPT

Agent systems

An agent system is a network of several logical units, each of which can independently solve a specific task. Such a task could be an online search, for example, or querying a weather forecast from a weather service. The agents in the orchestra can work both with and without AI support.

Agent systems can be used to solve tasks that require several complex sub-steps to solve. One challenge here is checking the partial results. After all, the agent system has to decide whether a subtask has been completed or not. It is also challenging to break a problem down into sub-problems. Thirdly, it is crucial to interpret the results of the individual agents correctly.

Agent systems can already be orchestrated via platforms such as n8n. However, a lack of programming knowledge can only be replaced to a limited extent by the capabilities offered by a black box. In this respect, the wheat will be separated from the chaff when it comes to previously unsolved tasks for which no suitable assistance APIs are available. Many non-developers are also increasingly relying on services with very low usage fees. In the medium term, this leads to dependency, higher costs, a lack of transparency and poorer results (see Black Box).

Consequences:

• New inventions and active ingredients, but also warfare agents, will be created with the help of AI.
• Humans are becoming less important, at least as far as thinking activities up to a certain level of intelligence are concerned.
• The automation of processes will advance massively.

Rise of autonomous systems

Autonomous systems, especially robots and drones, will make great progress. More and more AI skeptics will realize that robots are a threat to humanity. AI skeptics will keep inventing new reasons why AI is unlike intelligence and cite human intelligence as the benchmark.

The rise of robots will eliminate some jobs as early as 2025. This is likely to start outside Europe because that is where the manufacturers of these robots are usually based. In very few years, households of high earners, who do not have to be wealthy, will be using domestic robots.

Cyber attacks will increasingly target such autonomous systems in the future. Anyone who doesn't yet have a smart home but likes a thrill should automate their home as soon as possible.

Consequences:

• People are increasingly relieved of everyday and repetitive industrial tasks.
• Care for the elderly is relieved somewhat.
• War is increasingly being waged via drones (and digitally).
• Humans are becoming increasingly insignificant. The rise of robots is becoming more and more apparent.

AI divides society

All those who think of AI as witchcraft and deny AI the ability to intelligent behaviors are getting increasingly unlucky. Many see AI systems as stochastic parrots. They will despair that AI, which in their eyes is "only" a simulation of intelligence, will take over more tasks from humans.