reaction to recent releases on X, noting that one of the biggest‘ wow’ moments was moving from GPT-3.5 to GPT-4. Yet now, open-source models with only 2.6 billion parameters running locally on a handheld phone can outpace the original GPT-4 benchmark.
Deep learning in production and applied enterprise AI While consumer apps grab headlines with local edge models, the enterprise sector faces a different engineering challenge: how to run deep learning in production reliably, securely and cost-effectively.
Dwarak Rajagopal, who led AI Engineering and Research at Snowflake until July 2026, says the conversation shifted quickly from theoretical capabilities to the realities of systems engineering.
“ My team focuses on system efficiency – making sure companies get the most out of their compute power while keeping sky-high AI costs under control,” he adds.
To solve these systemic bottlenecks, engineering teams need to dig into lower-level framework optimisations and custom training protocols, rather than simply sending API calls to commercial frontier labs.
technologymagazine. com 155