Big Technology Podcast

The episode has genuine pockets of substance - the technical explanation of voice-model turn-taking limitations, the Codex-equals-Slack internal penetration claim, and the 230M weekly health-query figure are all non-obvious. But the conversation is padded with extended vision-casting, AGI platitudes, and repeated 'we're just at the beginning' framing that dilutes the signal-to-noise ratio.

“we basically have the same level of penetration now and usage as slack. Right. OpenAI is like an entirely Slack based company”

“The way that it works and the way that everyone's models work is that you basically chain together... you still have this problem of turn taking”

Most of the content is standard OpenAI positioning dressed in slightly fresh language; the 'interface melting away' and 'trust as earned, not granted' framings are genuinely interesting but not sustained. The contrarian moment on voice UX - framing turn-taking as humans contorting to machines - is the sharpest original thought, but it passes quickly and isn't developed into a deeper argument.

“turns again are so unnatural. This is the humans contorting ourselves to the machine and its limitations”

“And that's something we view as earned. It's not something that we can grant”

Greg Brockman is a co-founder and president of OpenAI - arguably the single most directly relevant practitioner alive for this subject matter. He speaks from genuine first-hand product and research experience, including internal metrics, chip program details, and personal health use-cases, not borrowed frameworks or secondhand observation.

“we've been investing in our own chip program now for multiple years and super exciting progress. We'll have more to announce, uh, actually pretty soon”

“We have our own network protocol that we've had to design, um, that we have people who look at every single layer of the stack”

There are several anchoring specifics - 230M weekly health users, Codex-Slack parity, 512M token context, the 1959 perceptron benchmark, and the 20-year mystery diagnosis anecdote - giving the episode real evidential texture. However, many of the biggest claims (revenue ramp, compute demand, differentiation advantage) are supported only by assertion and vague exponential gestures.

“there's about 230 million people each week who use ChatGPT for health queries”

“neural nets themselves were designed in like 1940s before computers. Right. As a model of maybe this is how the brain processes information. First hardware implementation was 1959 with the perceptron”

The host occasionally lands a sharp question - pushing on price cuts, pressing the Satya Nadella rivalry, asking directly about device strategy - but too often summarizes the guest's prior answer back to them as the next question, and lets evasive non-answers (on GPT 5.6, on Microsoft feelings) slide without meaningful follow-up. The format feels more like a friendly keynote interview than a probing conversation.

“Are you going to cut prices?”

“Okay, just to go back to the satya point one more time, uh, he's called Model's a commodity. He's trying to build his own frontier intelligence. He's telling potentially your customers, hey, you gotta come work with us”