Introduction to Hoptroff
WATCH NOW
“ As Gordon Moore said , If everything you try succeeds , you ’ re not trying hard enough ”
Back then , as now , availability of data drives what you can achieve .
In the 1990s , I specialised in extracting the best insights possible from small datasets , because that was all anybody had . With the Large Language Models , such as ChatCGP , and image creators such as DALL-E 2 , he who holds the data holds the power . And , even if the dataholders are willing to share data , they may not be able to do so for privacy reasons .
A company I was a director of , Replica Analytics – now acquired by Aetion Inc – specialised in creating synthetic datasets from real medical data so that AI models and software testing engineers could take advantage of it .
In terms of emerging trends , I contrast the AI world of the 1980s with the AI world of today . Back then , we had crazy , interesting ideas that have never been followed up on . Today , a few big players use a narrow branch of AI to extract value from the data they sit on . I think the new advances will be in revisiting those old ideas . Ones I ’ m particularly keen on are error estimation (“ ChatGPT , are you sure you are right ?”) and anomaly detection , where you could connect up an AI system to any input , it learns the patterns , and raises an alert when the data doesn ’ t fit the usual pattern .
technologymagazine . com 19