0:00 Salesforce AI Research just reduced the latency for voice-based AI by 316 times.0:06 That isn't an incremental improvement.0:09 It's a fundamental change in how conversational AI can work, moving it from a clumsy tool to something that can keep up with human speech.0:19 The other major signal today is in hardware.0:23 South Korean startup Rebellions just raised four hundred million dollars to build new AI chips.0:30 This confirms the silicon race is now a global, multi-front war, not just an Nvidia monopoly.0:37 Following that money, EPIC Microsystems raised twenty-one million dollars to tackle the growing power delivery problem in AI data centers.0:48 Without better power tech, all the new chips are just expensive heaters.0:53 On the software side, Google released its Agent Development Kit for Java, pulling AI development deeper into established enterprise ecosystems where Java still dominates.1:07 And Deccan AI raised twenty-five million dollars to focus specifically on post-training infrastructure...1:15 making enterprise AI more reliable after it's been deployed.1:19 Finally, we saw more specialized hardware emerge.1:23 Contec launched a new high-speed analog input card designed for testing semiconductor devices and EV motors—a direct consequence of AI driving demand in adjacent industries.1:37 And MiniMax released a multimodal server for generating text, speech, and video, though early reports suggest it’s a cost-versus-quality trade-off, achieving about ninety percent of the quality of top models for a fraction of the price.1:55 Let's go back to that Salesforce number.1:58 A 316x latency reduction.2:00 How?2:00 By changing the architecture.2:02 They call it VoiceAgentRAG, and it uses a dual-agent design.2:07 The problem with most voice assistants is they have to think before they speak.2:13 When you ask a question, the model has to go retrieve information, formulate an answer, and then generate the audio.2:22 That entire process creates an awkward, unnatural pause.2:26 Salesforce split the work.2:28 They created a "Fast Talker" agent that only checks a local semantic cache.2:34 It handles the immediate, low-latency part of the conversation.2:39 At the same time, a "Slow Thinker" agent goes and fetches new documents or data in the background, asynchronously.2:48 This decoupling is the entire trick.2:51 The AI can start giving you an initial response in under 200 milliseconds—the budget for making conversation feel natural—while simultaneously preparing the deeper, more complex information.3:05 It's the difference between an AI that has to stop and think, and one that appears to think while it talks.3:14 This directly addresses the AI readiness gap.3:17 The problem isn't just that companies feel unprepared to deploy AI...3:22 it's that many AI tools, especially voice tools, are still too frustrating for anyone to actually use.3:30 All the funding for custom chips and new platforms is irrelevant if the final experience is broken.3:38 The focus is shifting from simply building larger models to solving the practical constraints of power and latency.3:47 The real work is no longer about making the machine think.3:51 It’s about making it think without that awkward pause.
