Sign In

Carlos Escapa

Global Business Development, Data & AI/ML Solutions

Carlos Escapa is a seasoned professional with a passion for building global businesses in the technology sector. With a strong background in AI/ML, cloud computing, and open source, he thrives in multicultural environments that prioritize accessibility and simplicity in technology adoption. Carlos has a rich history of living and working in various international locations, including Washington, Tokyo, Paris, and the San Francisco Bay Area.

Highlights

Aug 24 · linkedin

Gartner has put out a hype cycle riddled with mistakes

  1. GenAI is already in the trough
  2. Reinforcement learning is well past the trough into the plateau - AlphaGo happened in 2015
  3. Ditto for Federated ML, used by Google in full production to train the Gboard since 2016
  4. Can autonomous agents really be trailing my favourite bullshit technology, "Artificial General Intelligence"?
  5. AI Supercomputing is here today, to wit: power shortages in Northern Virginia, Oregon and Ireland
  6. Superapps have existed in China for at least 7 years

https://lnkd.in/dZYT-aRA

Aug 10 · linkedin

The tempering of GenAI hype Why are analysts so slow and cautious to call it out?

David Linthicum brings up a point that has been evident for well over a year - CIOs are nowhere near as enthusiastic as Satya Nadella or Sam Altman about the value of LLMs.

Or "Artificial General Intelligence" for that matter.

The battleground, as my esteemed colleague Julien SIMON never tires to point out, is Small Specialised Models, and not the bloated, inefficient and planet-eating LLMs.

SSMs are being deployed in production, generating value quietly and efficiently.

They solve specific problems - semantic search, assisting document authoring, triaging, prioritising, expanding self-help systems. Lots more.

90% of LLM projects are withering in never-ending POC cycles, waiting for magic dust. The other 10% are deployments far away from areas that require precision and compliance.

We're really back to what we have long known, which is that Machine Learning is best applied to specific problems that can be solved with proprietary data.

"It's the data stupid" writ large.

#aiisnotintelligent

Carlos Escapa
Add to my network

Location

San Mateo, California, United States