Will a new lab create a top-performing AI frontier model before 2028?

Will a flagship (>60T training bytes) open-weights LLM from Meta which doesn't use a tokenizer be released in 2025?

Will models be able to do the work of an AI researcher/engineer before 2027?

10GW AI training run before 2029?

Will a GPT-3 quality model be trained for under $10.000 by 2030?

Before 2028, will there be enough inference capacity to generate 30T frontier model tokens per day?

Will a model costing >$30M be intentionally trained to be more mechanistically interpretable by end of 2027? (see desc)

Will there be a more sample-efficient pretraining algorithm than next token prediction for NLP before 2027?

Will OpenAI release a tokenizer with more than 210000 tokens before 2026?

Before 2028, will any AI model achieve the same or greater benchmarks as o3 high with <= 1 million tokens per question?