In this episode of the Agent Workforce Podcast, Kristiina and Niko reunite to discuss the latest in LLM trends, model chaos, and real-world AI adoption. From Karpathy’s Y Combinator summer school talk to what’s actually inside a system prompt, we cover a lot of ground — including energy use, pricing shifts, and the surprising rise of OpenEvidence as the go-to tool for U.S. physicians.

Plus: insights into Sam Altman’s comments on AI resource use, GPT-4 pricing shifts, and why evals are actually the unsung heroes of scalable AI solutions.

🔍 Topics include:

• Karpathy’s Y Combinator summer school talk: “software is changing again”

• The evolving GUI of LLMs & user autonomy

• How much energy ChatGPT really uses

• OpenEvidence is becoming the “search bar” for U.S. doctors

• 03 model price drops and selection confusion

• Prompting at different levels (system, user, custom GPTs)

• How Agent Workforce agents are built under the hood

• The role of evals in building trustworthy agents

• A quick summer sign-off and a shoutout to ⁠agentacademy.ai⁠