On: Microsoft's paper on OpenAI's GPT-4 had hidden information

[singularity2001]

there is no hard line but gpt4 solves some tests 90% correctly where 3.5 only achieved 10%