CarahCast: Podcasts on Technology in the Public Sector · Carahsoft

Inside Intel: Shaping the Future of Compute

·32 min·2 clips

Agents do most of the work on CPUs, not GPUs, Darren says.

1. CarahCast: Podcasts on Technology in the Public Sector features a conversation about Intel, public-sector AI, and data center power use. 2. Scott Robon hosts the episode as Kerasoft Consulting CTO, and Dr. Darren Pulsifer joins as Intel’s Chief Solutions Architect, which matters because both speak directly to public-sector infrastructure and AI deployment. 3. The episode asks what AI is for when cost, energy, and data quality become the limiting factors. 4. Pulsifer says November 20, 2022, the day ChatGPT launched, mattered because accessibility and ease of use changed who could work with AI. 5. He compares ChatGPT’s impact to the iPhone and the mouse, saying the technology had existed before but became widely adopted when it became easier to use. 6. He says Sam Altman made ChatGPT highly accessible, which let more people start working with AI without a data-science team. 7. Pulsifer argues that money flowing into AI after ChatGPT created more innovation, and he says much more is happening beneath the surface through API access and developer use. 8. The conversation shifts to energy, where Pulsifer says ChatGPT5 has a power envelope 10x that of GPT-4.0. 9. He says the industry cannot keep scaling energy demand at that rate because there is not enough power to support it. 10. He points to small language models on AI PCs and neuromorphic processing units as a way to move from kilowatts to watts. 11. Pulsifer says small language models are useful for processing unstructured data into structured outputs, especially when the task does not require creativity. 12. He gives a financial example using SEC 10-Ks and 10-Qs, where a model can extract revenue streams from PDFs with a constrained prompt. 13. He says the U.S. Census Bureau is using similar methods to help analysts work through unstructured survey data and structured data faster. 14. One memorable claim is that no one should be training their own models anymore because so many trained models already exist. 15. He says context engineering and retrieval augmented generation can supply the needed context without expensive model training rigs. 16. Pulsifer says collaborative work between humans and AI is already happening at the U.S. Census Bureau, where small context databases help the model adapt to a worker’s interaction style. 17. The tone is practical and opinionated, with Robon pressing for examples and Pulsifer answering with specific workflows, numbers, and vendor names. 18. The format stays conversational, moving from career background to AI economics, then into data curation, training strategy, and career advice. 19. Best for public-sector IT leaders, data engineers, and AI-curious students. 20. Skip it if you want abstract theory without implementation detail.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from CarahCast: Podcasts on Technology in the Public Sector