Clarify the target: define AGI in measurable terms
Identify the exact capabilities missing from last year’s best systems
Collect all model releases, papers, benchmarks, and evals from last year
Reproduce the strongest last-year baseline models
Audit data quality, training recipes, and inference settings
Compare last-year systems against current AGI requirements
Find the smallest set of changes that closes the gap
Improve reasoning, memory, planning, and tool use
Add long-context and retrieval capabilities
Strengthen multimodal understanding if needed
Train on higher-quality, more diverse data
Use better synthetic data generation and filtering
Apply stronger post-training and alignment methods
Run rigorous evaluation on hard, unseen tasks
Iterate on failures until performance is robust
Scale compute, data, and model architecture where necessary
Validate safety, reliability, and generalization before deployment