Top 5 trends in Artificial Intelligence for 2023 ๐Ÿ“ˆ

Reflections with SaugaTalks on five defining shifts shaping enterprise AI adoption: generative models, data governance, explainability, spatial computing, and low-code ML.

Generative AI Explainable AI AI Governance Data-Centric AI

In the last week of 2022, I was having a conversation withย Irene Lyakovetskyย fromย SaugaTalksย about how incredible of a year it was with some really cool AI applications like Chat GPT, DALL E V2, PaLM, and so many more.

As we discussed if AI can be sentient and would AI-powered movies be that enjoyable, we drew out a list of the top 5 trends we are excited about in the upcoming year ย ๐Ÿ‘€

๐Ÿ“Œ ๐—š๐—ฒ๐—ป๐—ฒ๐—ฟ๐—ฎ๐˜๐—ถ๐˜ƒ๐—ฒ ๐—”๐—œ that mimics creativity, one of the most unique skills humans possess.

๐Ÿ“Œ Comprehensive ๐——๐—ฎ๐˜๐—ฎ ๐—ฃ๐—ฟ๐—ผ๐˜๐—ฒ๐—ฐ๐˜๐—ถ๐—ผ๐—ป ๐—Ÿ๐—ฎ๐˜„๐˜€ ๐—ฎ๐—ป๐—ฑ ๐—ฅ๐—ฒ๐—ด๐˜‚๐—น๐—ฎ๐˜๐—ถ๐—ผ๐—ป๐˜€ as curation and consumption of native/synthetic data amplifies with a focus on data-centric approaches to AI.

๐Ÿ“Œ ๐—˜๐˜…๐—ฝ๐—น๐—ฎ๐—ถ๐—ป๐—ฎ๐—ฏ๐—น๐—ฒ ๐—”๐—œ to steer clear of irreversible damages that can be caused by applications of AI risk-prone fields like healthcare, defense, and legal.

๐Ÿ“Œ ๐— ๐—ฒ๐˜๐—ฎ๐˜ƒ๐—ฒ๐—ฟ๐˜€๐—ฒ with its mind-boggling promise to transcend the physical layer and provide a โ€˜feeling of physical presenceโ€™ despite being in a virtual world.

๐Ÿ“Œ More sophistication in ๐—Ÿ๐—ผ๐˜„-๐—–๐—ผ๐—ฑ๐—ฒ/๐—ก๐—ผ-๐—–๐—ผ๐—ฑ๐—ฒ ๐—ฆ๐—ผ๐—น๐˜‚๐˜๐—ถ๐—ผ๐—ป๐˜€ to boost the adoption of AI in multifarious domains and help build an MVP quickly.

SaugaTalksย also released a video with additional details/context, check that outย ๐ŸŽฌ

Continue Reading
Aug 2022

Practitioner notes on foundational machine learning papers shaping modern representation learning and production recommendation systems.

ML Research & Foundations
Nov 2026

Catching silent CPU fallbacks, hidden Hugging Face host syncs, and XLA recompiles in PyTorch on Cloud TPU v6eโ€”cutting LoRA step time 2,537ร— (47.3s โ†’ 18.6ms) alongside upstream PRs to transformers & peft.

Fine-Tuning & Compute
Sep 2026

How Proof of Work uses Gemini's agentic video understanding to evaluate software engineering process, debugging discipline, and hypothesis iteration from continuous screen recordings.

Agentic AI & Evaluation