Stephen Wolfram has learned a variety of lessons, some of which can be inferred from the provided context. Here are some of the surprising lessons along with relevant timestamps:

1. The Workflow of Language and Code (00:49:00 - 00:49:29):
- Wolfram has learned about a new workflow that involves humans mumbling something, after which a large language model produces a fragment of Wolfram Language code. This code can then be run to see if it produces the correct result, which is a process that involves both human and machine intelligence.

2. The Nature of Truth and Computation (02:14:30 - 02:14:55):
- The impact of large language models like GPT on society has led to lessons about the nature of truth, especially in relation to computational systems like Wolfram Alpha that use symbolic reasoning.

3. Writing Code with Natural Language (00:42:00 - 00:42:28):
- Wolfram has been considering the translation of natural language to computational language since the early days of Wolfram Alpha and has been surprised by the progress in this area over time.

4. Computational Irreducibility (00:18:30 - 00:18:59):
- The lesson of computational irreducibility seems to be important for understanding complex systems, including the universe itself, which operates on a scale and complexity beyond straightforward computation.

5. Ethical Implications (01:53:30 - 01:53:57):
- Wolfram reflects on the implications of one's work, likening the situation to the Manhattan Project, where the outcomes of significant technological advancements can lead to profound ethical considerations.

6. Natural Language and Computation Success (00:41:30 - 00:41:59):
- Wolfram Alpha's success in interpreting natural language queries has been a surprising lesson, reaching a high success rate over the years, which could be due to a feedback loop where more successful queries are used more often.

These lessons indicate an evolving understanding of computation, language, and the interaction between humans and machines. Wolfram's insights span the technical, philosophical, and ethical dimensions of his work in computational science.