Be it data science or machine learning. People start asking: Has ML run its course?

All the people write about web3 & NFTs now. But this is just the field maturing.

Here's how we do the real exciting work in DS, ML & AI now!

πŸš€ Innovations speed of ML

It may seem like the innovation speed of ML is slowing down. But is it really?

In 2014 we got Generative Adversarial Networks. In 2017 we got Transformers. In 2020... anyways ignore that.

These technologies are maturing with GPT-3 and CLIP!

πŸ›  Operations of ML

We are slowly figuring out how to build maintainable end-to-end pipelines.

  • Data Engineering
  • Data Validation & Schemas
  • Automated training
  • Concept drift

MLOps is very new and hard but exciting

i.e. at ECMWF we need to make ML work with Fortran!

πŸ’₯ But the failures of data science!

We hear about business failures and companies scrapping data teams.

But for every story of failure, I see 3 successes. Data-first companies make ML/DS work

ML/DS truly drives business value.

My chat w/ Candost Dagdeviren:

βš™ Tools are consolidating

Every day, I see a new awesome library that makes ML and DS easier.

Just follow Philip Vollet or subscribe to my newsletter.

  • Pandas profiler automates EDA
  • Pandera validates data frames
  • Keras/Lightning make deep learning easy

πŸ“‰ The AI hype is over

Tesla & self-driving is well into the "valley of despair" of Gartner.

Deep learning is over the hype peak.

The hype is over but now we can work with realistic expectations and build useful real-world machine learning and data science applications


  • Innovation speed of AI is still going
  • MLOps is becoming more important
  • Data-first companies succeed at ML & DS
  • Awesome tools make life easier
  • The AI hype is slowly dying down

but it doesn't feel like an AI winter.

ML is applied successfully everywhere. Just maybe not everywhere locally.