Post by account_disabled on Sept 10, 2023 9:46:54 GMT
It’s been a long journey leading up to the creative-writing AI of today, starting in the halls of MIT with industry giants like Marvin Minsky and John McCarthy of the legendary Media Lab who laid a lot of foundation in the 1960s and 70s, but with disappointing results, cutting into credibility and leading to what we now call the “AI Winter”. It turned out to be greatly a matter of the hardware not being ready.
The concept of the personal data assistant popped up over the years such as the much maligned but forward-thinking handwriting recognizing Apple Newton in 1993 and the first popularly successful PDA, the US Robotics PalmPilot in 1997, paving the way for today’s AI-hardware equipped smartphones.
Google’s foundational PageRank from 1998 is a form of AI in that it is a “machine learning” algorithm. Google pushed a series of aggressive “invisible” product advancements such as better Phone Number List Google Maps, quietly improving quality against a backdrop of boring. There were sexier promising starts along the way, such as 2011 when they first rolled out Voice Search in the Chrome browser, then in 2014 when voice search hit mobile Android phones. The AI reality was underwhelming, but anticipation was being built.
In 2015, Google announced that new AI-powered search infrastructure called RankBrain, followed by advancements that were labeled neural matching, BERT and MUM, all of which are language-processing precursors to what took the world by storm in November of 2022 when OpenAI released a product built on a seminal Google paper published Thursday, August 31, 2017 on novel new neural network architecture for understanding language.
Transformational transformers
Only a year after Tay, Google released a paper on the Transformer, a new type of neural network that was able to do machine translation better than anything that had come before. It’s the “T” in GPT, and has made new machine learning output considerably more compelling than Clippy or Tay, with the simple trick of predicting what’s statistically most likely to be typed next—a profoundly deeper thing than it seems at first glance.
This caught Microsoft’s attention, who invested $10 billion in July 2019. Several earlier GPT versions available through the API-only were released and had many developers playing in a playground, but it failed to capture the public’s fancy, behind a login and not yet following the chat paradigm as it was.
Nov. 30, 2022: ChatGPT and the fastest new service adoption rate in history
The first version of OpenAPI’s GPT for the general public, ChatGPT, was launched November 30, 2022. The original ChatGPT release was based on GPT-3.5. A version based on GPT-4 was released on March 14, 2023 (fast-forward in the timeline) right as Microsoft announced their intention to power The New Bing with the latest version, 4.5.
While the period between November 30, 2022 and March 14, 2023 was only 3.5 months, it was a period of intense experimentation and learning for Microsoft and the public, with the now famous fastest adoption-rate of any new online service in history. Things are moving so fast now, it’s time to look at the timeline.
The concept of the personal data assistant popped up over the years such as the much maligned but forward-thinking handwriting recognizing Apple Newton in 1993 and the first popularly successful PDA, the US Robotics PalmPilot in 1997, paving the way for today’s AI-hardware equipped smartphones.
Google’s foundational PageRank from 1998 is a form of AI in that it is a “machine learning” algorithm. Google pushed a series of aggressive “invisible” product advancements such as better Phone Number List Google Maps, quietly improving quality against a backdrop of boring. There were sexier promising starts along the way, such as 2011 when they first rolled out Voice Search in the Chrome browser, then in 2014 when voice search hit mobile Android phones. The AI reality was underwhelming, but anticipation was being built.
In 2015, Google announced that new AI-powered search infrastructure called RankBrain, followed by advancements that were labeled neural matching, BERT and MUM, all of which are language-processing precursors to what took the world by storm in November of 2022 when OpenAI released a product built on a seminal Google paper published Thursday, August 31, 2017 on novel new neural network architecture for understanding language.
Transformational transformers
Only a year after Tay, Google released a paper on the Transformer, a new type of neural network that was able to do machine translation better than anything that had come before. It’s the “T” in GPT, and has made new machine learning output considerably more compelling than Clippy or Tay, with the simple trick of predicting what’s statistically most likely to be typed next—a profoundly deeper thing than it seems at first glance.
This caught Microsoft’s attention, who invested $10 billion in July 2019. Several earlier GPT versions available through the API-only were released and had many developers playing in a playground, but it failed to capture the public’s fancy, behind a login and not yet following the chat paradigm as it was.
Nov. 30, 2022: ChatGPT and the fastest new service adoption rate in history
The first version of OpenAPI’s GPT for the general public, ChatGPT, was launched November 30, 2022. The original ChatGPT release was based on GPT-3.5. A version based on GPT-4 was released on March 14, 2023 (fast-forward in the timeline) right as Microsoft announced their intention to power The New Bing with the latest version, 4.5.
While the period between November 30, 2022 and March 14, 2023 was only 3.5 months, it was a period of intense experimentation and learning for Microsoft and the public, with the now famous fastest adoption-rate of any new online service in history. Things are moving so fast now, it’s time to look at the timeline.