After training nearly every major foundation model, we’re seeing critical shifts in how models are built, trained, and brought to life.
1. High quality pre-training data
As models become more sophisticated they demand more refined datasets, and accuracy and completion matter in every single response.
2. Shorter training sprints
A major shift in the industry is the move from long training cycles to shorter, more iterative processes.
3. Specialized and diverse data
Rather than training on general datasets, researchers are looking at fine-tuning models with highly specialized and diverse data.
4. Chain-of-thought reasoning
A key evolution of chain-of-thought (CoT) reasoning has been increased visibility into how models reason step by step.
5. Agentic products
Emerging as a major trend in AI, agentic products shift the focus from simple one-shot responses to multi-step, orchestrated workflows where models can call other models and tools to complete complex tasks.
2025 trends & insights
Get your free report with insights from Invisible's CEO, Matt Fitzpatrick
By clicking "Accept", you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.
