Home | Power BI | Excel | Python | SQL | Generative AI | Visualising Data | Analysing Data
The shortest history of Generative AI
Generative AI seems to have arrived suddenly but in fact it has a long gestation. Here are a few moments of AI history.
- 1956, Artificial Intelligence (AI) was born with the ambition to create intelligent machines at the Dartmouth Workshop
- 1990s, Machine Learning (ML), statistical learning from data and predictions.
- 2000s, commercial use of early single purpose Generative AI: Google Translate (2006), Siri (2011), autocomplete when texting on a phone
- 2017, deep learning, an ML technique inspired by the wiring in our brains takes off. In deep learning, “neurons” are connected into layers and the weights of the connections between neurons can be adjusted so that the neural net can “learn”.
- 2021, an extension of the neural net approach using a new “transformer” architecture yields results.
Given that long history, what is the reason for the recent excitement? These LLMs have very recently become much more powerful and capable. For example, in 2023, OpenAI released ChatGPT-4 which can do some very impressive things:
- score more highly on the SAT, a US-based university entrance exam, better than 90% of people,
- pass some university level exams in law, medicine,
- set out arguments in favour and against a certain thing e.g. vaping,
- play an expert role e.g. act a a Python expert and write code,
- write a job description.
Anybody can use ChatGPT and with a few hours training can use it very effectively - no coding proficiency or technical skills are required.
Because of these factors, ChatGPT took only 2 months to reach 100m users, compared to 9 months for TikTok and 70 months for Uber.