AI Trends for 2025 (AI column)
I saw it recently. seven-minute video Martin Keen, IBM Fellow, who for many years at IBM proved himself as an inventor (more 400 patent)He is a technical supervisor, researcher in AI, and content creator on the IBM Technology channel. In this short video, he talks about his vision of AI trends. 2025 year. Xatachi did that last year, too. video 2024 yearSo you can go and compare how well his predictions hit the spot. But let’s get back to the forecasts for this year.
- Agentic AI This is an important trend for creating multi-step plans and their execution using available tools. However, while modern models are experiencing difficulties with logical thinking and execution of complex plans.
- Inference time compute Models already use more time during inference time, improving logical output. This allows you to customize and improve the system without the need to train/tune the basic model.
- Very large models - This year we're supposed to break through. 1 a trillion parameters in foundational models
- Very small models Smaller models that can work on laptops and phones will be actively developed.
- More advanced use cases AI will improve customer service, IT operations and automation.
- Near infinite memory Modern AI models have contextual windows of millions of tokens. Then there will be more and, for example, conditional chatbots will be able to store and use all available information on the user.
- Human in the loop augmentation Chatbots may outperform doctors in diagnosis, but paired with experts, they can be more effective. This requires better systems to integrate AI into workflows. For example, we actively integrate our Nestor into all processes around SDLC. (software development lifecycle) Increase the productivity of engineers
#AI #ML #Trends #Software #Engineering