About this episode Tune into AI, Unfiltered to get your daily dose of critical insights into the rapidly evolving world of artificial intelligence. We cut through the hype to bring you unbiased analysis of new models, product launches, research breakthroughs, and funding rounds, highlighting what truly impacts the AI landscape. This podcast equips you with the knowledge to discern signal from noise, helping you understand the real advancements shaping the future of AI.
0:00 google just released gemini four hydrogen pricing input at 1.50 per million tokens while claiming it edges out their frontier argon model on the artificial analysis index in episode two thirty we tracked the gemini four carbon rollout and anthropix public debut now the cycle is pivoting from raw frontier scale to cost discipline and execution speed here's what moved0:25 across the market ibm and microsoft launched imagine enterprise connecting over 6,000 ibm consulting agents directly into azure infrastructure for regulated industries openai rolled out daybreak red team and defense factory deploying automated penetration testing and patch generation across corporate code bases in funding startup marindle secured $1,000,000,000 in a series a led by kleiner perkins and clear lake capital0:54 to automate model pre training with agent swarms luminal raised $23,000,000 to target nvidia's cuda software moat by optimizing inference on mixed hardware on the academic side mit researchers published a framework in the proceedings of the national academy of sciences that trains climate emulators on targeted synthetic data beyond that paper verified foundational breakthroughs were absent today so where1:20 is the actual signal well look past the marketing claims and watch where capital is digging trenches google claims hydrogen runs 2.5 times faster than gemini 3.8 flash offering four discrete thinking levels so buyers only pay for the reasoning depth they need that is not an attempt to redefine general intelligence that is an enterprise product engineered to make1:46 autonomous agentic loops economically viable when an agent burns millions of tokens debugging a repository latency and token pricing determine whether the workflow ships or gets canceled look at openai's security launch and you see the exact same commercial pressure daybreak red team scans web applications for attack vectors while defense factory generates and validates code patches before engineers merge2:12 them openai leadership framed frontier models as nonnegotiable enterprise defense okay that rhetoric sounds dramatic but strip away the threat modeling and examine the mechanics it is vendor capture once an organization trusts an external model to scan proprietary repositories and draft security patches switching infrastructure providers becomes a massive operational risk anyway2:41 here's the thing you need to track as this quarter plays out2:46 the narrative of exponential capability leaves is hitting the reality of unit economics frontier training still attracts capital as miranda's billion dollar round demonstrates but on the ground the decisive battle has shifted from benchmark bragging rights to enterprise plumbing compliance guardrails and sustainable inference costs the vendors winning today aren't expanding what models can conceptualize they're grinding down the3:15 price of letting them act
