As someone who contributed to some of those waves in terms of core tech, got various niche recognitions for it, and collaborated+employed folks here, I don't think it died but continues to evolve. (The Graphistry team helped with GPU dataframe / viz movement, apache arrow, etc, and pay our bills with folks needing to answer tricky graph questions such as on linking events or people.)
Much of the shift IMO is about who & goal, and quite exciting: gen AI is unleashing a lot of old & new ideas.
Jupyter, then streamlit, got big. In parallel, ML and now AI. So less on bespoke viz on large data, often with superficial levels of analytics. Now more on ML & AI pipelines to both understand more, and in the generative era, do more. Amazingly, you don't need to be a js wunderkind either, almost anyone with a database can do it.
Nowadays as our team is launching Louie.ai (genAI notebooks, dashboards, & pipelines), it is in part using gen AI to make all our years of GPU/graph/ai tech easy for more of our users. But the best viz is no viz, and the second best is just an answer or a list, and with genAI, one you can ask questions to. So viz is not a goal but a means, and part of a broader process like investigation and automation. We are having to start over a LOT, and so it is not dead, but changed.
Funny enough, one of the last public talks I gave before starting Graphistry all those years ago was at Strangeloop (10 years ago?) where I was describing program synthesis as the future of visualization, and demoing prompt engineering with SAT-solver-era tools. However, that wasn't powerful enough back then for what BI users needed in practice, so the program synthesis leg of our vision just wasn't practical until gpt-3.5/gpt-4 came out 2 years ago. As soon as that happened, we resumed. Building louie.ai has been incredibly stimulating and exciting now that we can finally do it all!
As someone who contributed to some of those waves in terms of core tech, got various niche recognitions for it, and collaborated+employed folks here, I don't think it died but continues to evolve. (The Graphistry team helped with GPU dataframe / viz movement, apache arrow, etc, and pay our bills with folks needing to answer tricky graph questions such as on linking events or people.)
Much of the shift IMO is about who & goal, and quite exciting: gen AI is unleashing a lot of old & new ideas.
Jupyter, then streamlit, got big. In parallel, ML and now AI. So less on bespoke viz on large data, often with superficial levels of analytics. Now more on ML & AI pipelines to both understand more, and in the generative era, do more. Amazingly, you don't need to be a js wunderkind either, almost anyone with a database can do it.
Nowadays as our team is launching Louie.ai (genAI notebooks, dashboards, & pipelines), it is in part using gen AI to make all our years of GPU/graph/ai tech easy for more of our users. But the best viz is no viz, and the second best is just an answer or a list, and with genAI, one you can ask questions to. So viz is not a goal but a means, and part of a broader process like investigation and automation. We are having to start over a LOT, and so it is not dead, but changed.
Funny enough, one of the last public talks I gave before starting Graphistry all those years ago was at Strangeloop (10 years ago?) where I was describing program synthesis as the future of visualization, and demoing prompt engineering with SAT-solver-era tools. However, that wasn't powerful enough back then for what BI users needed in practice, so the program synthesis leg of our vision just wasn't practical until gpt-3.5/gpt-4 came out 2 years ago. As soon as that happened, we resumed. Building louie.ai has been incredibly stimulating and exciting now that we can finally do it all!