GPT-5 podcast mentions

Quote

be running on Google Cloud. The top layer is the agent layer in the browser and the workspace. Google is, you know, the only player that's hitting all three credibly right now. Anthropic is obviously crushed But they're not making their own chips, right? And so Google has a big advantage when they own the entire stack. Microsoft is really heavy on open AI at layer two, but really weak on silicone. Amazon is betting the farm on anthropic. Nvidia is a chip layer and doesn't touch the application layer. Google is basically the only company really doing the full stack. So I think the strongest argument against this is that Gemini still hasn't closed the gap with GPT 5.4 or Claude Opus 4.0. seven on the benchmarks that consumers actually care about. And so I think, you know, maybe those TPU numbers are Google's own. And you could also say, you know, until like an independent lab pod publishes real world training runs on the um eight T and on the eight I the three X claim that they've given is marketing and I mean I don't wanna I don't wanna uh doubt Google too much but But you have to be uh you have to be fair. There is a lot of these AI companies and Google's been known to do this in the past, um, to make big claims on marketing and it it doesn't always stack up in the real world. Here's why I still think even with all of that, um Google is in an incredible position right now. The…

Quote

be running on Google Cloud. The top layer is the agent layer in the browser and the workspace. Google is, you know, the only player that's hitting all three credibly right now. Anthropic is obviously crushed But they're not making their own chips, right? And so Google has a big advantage when they own the entire stack. Microsoft is really heavy on open AI at layer two, but really weak on silicone. Amazon is betting the farm on anthropic. Nvidia is a chip layer and doesn't touch the application layer. Google is basically the only company really doing the full stack. So I think the strongest argument against this is that Gemini still hasn't closed the gap with GPT 5.4 or Claude Opus 4.0. seven on the benchmarks that consumers actually care about. And so I think, you know, maybe those TPU numbers are Google's own. And you could also say, you know, until like an independent lab pod publishes real world training runs on the um eight T and on the eight I the three X claim that they've given is marketing and I mean I don't wanna I don't wanna uh doubt Google too much but But you have to be uh you have to be fair. There is a lot of these AI companies and Google's been known to do this in the past, um, to make big claims on marketing and it it doesn't always stack up in the real world. Here's why I still think even with all of that, um Google is in an incredible position right now. The…

Quote

Um it's been a it's been a year plus project. Um and the reason that I wasn't in that much of a FOMO because I wanted to get this right. This is was intended for um platform builders and enterprises that want to um to create their own um agendic platforms. So I'm not really concerned that uh when I started there we were GPT-4 and now we're GPT 5.2 thinking high max or whatever it may be tomorrow uh because the models uh are treated as as commodity and the uh the system is going to use the models based on the models that you've allowed it, the API keys that you provided. based on the type of task. So if you say, okay, creative work, I want you to use Opus, uh coding work, I want you to use codex. If you need to just do a quick and dirty cleanup or whatever, just use a minimax or uh or something, GLM or something that's lower cost. Yeah, so it has access obviously agent skills and knowledge bases and SOPs. Uh there's no the concept of a workflow doesn't exist there, uh, because I think that it's an more into uh uh to what a gentic means. But we do have the concept of SOP. So we have standard operating procedures. Essentially it's a template.

Quote

Physical shortcut cuts latency by up to five times. It is a necessary biological evolution for AI hardware. Oh, and just to add a layer of corporate intrigue to the hardware. Space. Google is reportedly working with Marvell to help design future chips because they want to break their long-term reliance on Broadcom. They want full control. The hardware is physical. Basically splitting and evolving just to accommodate the relentless pace of the software model releases. And this week the software barrage was overwhelming. Let's run through the timeline. OpenAI released GPT-5.5. This is a critical release because OpenAI explicitly designed GPT-5.5. to complete complex work with significantly less human direction. It's acting on its own. They claim it can act independently on a user's command across disparate tools like email clients, spreadsheets, and calendar apps. We are moving from the paradigm of a digital assistant that waits for prompts to an autonomous worker that executes workflows. They also released ChatGPT images two point zero, adding a specific reasoning mode designed to interpret complex scientific charts. Which is very advanced vision. And highly notified. Notably, they introduced ChatGPT for clinicians. This is a free tool for verified U.S. health workers running on a customized GPT 5.4 model. According to the The internal benchmarks it scored a 59.0 on the Health…

Quote

Same source. OpenAI's internal memo reveals a shift to enterprise platforms, with business revenue hitting 40% and aiming for half by year-end. Intensifying competition with anthropic. These cases highlight predictive analytics and finance, and natural language processing for marketing. Implementation challenges include governance lagging adoption. PwC's 2026 predictions note, urging top-down strategies with talent and change management. Integration demands data quality and edge AI for real-time decisions, while technical needs like OpenAI's GPT 5.4 enable agentic workflows at scale. Practical takeaways. Start with one high-impact process, like supply chain forecasting. Measure ROI via productivity gains of up to 40% from AI. automation, talent 500 reports, and pilot integrations with existing systems using multimodal models. Looking ahead, trends point to vertical AI. Human AI collaboration and cybersecurity defenses per SDG Group and Means CEO's April 2026 analysis, promising hyper-personalized Experiences, but requiring ethical oversight. Thank you for tuning in listeners. Come back next week for more. This has been a Quiet Please production. For more, check out the video. Check out QuietPlease.ai.

Quote

day-to day. It means a radical refocusing of resources. We're seeing evidence that they're pulling bringing people and money away from let's call them side projects or longer term ad revenue ideas. So, stop chasing shiny objects. Focus on the core product. That's the mandate. All hands on deck for core chat GPT improvements and the GPT five line. Right. Now to be clear, open AI is a little bit more than a little bit of a still incredibly strong. GPT 5.1 still leads in a lot of day-to-day conversational stuff. and it's deeply embedded in developer workflows. But the internal message is clear. What makes us think they can solve these fundamental problems now? under this kind of pressure? That really is the million dollar question. The pressure is immense. But what's shocking everyone is the sheer speed of Google's iteration. It feels like Google's unified architecture is just faster to tune and deploy. that way. So the urgency for OpenAI is to prove GBT5 isn't just a small step up. But a massive leap in reliability that restores that enterprise confidence. Next battlefield distribution. If model performance is becoming a slug factor,

Quote

It can't be successful and I'm not gonna give any predictions on the future, but uh I think that that the general trend of models becoming more integrated and more modalities becoming integrating the models is something that's gonna persist in the future. Sure. The things that are currently up to a user to differentiate. um the amount of thought, you know, image capability, things like that. You know, we've seen like in OpenAI, they've kind of abstracted away from that with the GPT-5 model. And it's the same thing in the Gemini app. If you use the Gemini app, then also uh there's one experience you can access uh nanobanana. in that app. So it's not um a a sort of a drop down choice there. It's a drop down for Okay. And is that did that change in the several weeks since it was launched. I remember Initially seeing like the the little banana thing, but I wasn't really sure or whether that was a discoverability tool or that was like a functional uh switch. The the banana you see in the app is a discoverability tool. Um, but since we launched uh at the end of last month it's been uh integrated into the Gemini app experience. Talk a little bit about the the the project as a whole, like

Quote

And a feedback layer, which determines how the first time we have to do that. system can improve over time, whether the results of the execution are verified, and whether each failure is recorded and transformed. So that's evaluation and verification and tracing and observability. And there is increasing evidence out there of the power of harnesses. Blitzy, who has over the last year been a frequent sponsor and collaborator on the show, recently released a sixty five. 66.5% performance on Suitebench Pro. That is much higher than, for example, GPT-5.4's 57.7. Now, effectively, Blitzy's whole thesis. Could be reframed as being that the harness layer, the agent scaffolding, the orchestration, and the context infrastructure wrapped around the foundation models, can unlock bigger performance gains. than the model themselves. One of the key things that they found when auditing their performance versus GPT five point four is that in many cases, GPT-5.4's failures weren't cataloged. Catastrophic. It got close on every problem, but missed intricate details and corner cases. When Blitzy succeeded on those same tasks, it succeeded because its knowledge graph gave its Agents, deep code-based context that a raw model doing a single pass couldn't match. Langchain has also recently been writing about how they've been improving agent performance with harness engineering. As well. Nicholas Cha…

Quote

Gemma 4 is like almo if like it's almost like GT. But it It's local and small and fast and you can run it on anything and it's got a big context and it's completely free. uh it's like it's clearly not state of the arts, right? I mean like it's like GPT 5 level on some benchmarks. Uh so you know it's like if it was like a it was a state of the art How do I put this? It feels like slightly better than GPT. Right? Like it is um like if this came out a year ago, uh, this would be the state of the art model. um that like everybody would want to use. But the difference is Is is uh a lot of people right now when they try to run the models locally on their computer, they need to spend like they need to decide all right do i want to buy a car or do i want to buy a Enough high priced GPUs and memory and um power in order to run one of these models, right? Like if you try to run

Quote

somewhat even sometimes easier to integrate uh because we come into it with a firm opinion. on what the sort of uh best way of using it looked like. And so Um some folks that we work with actually really appreciate that we come into it with an opinion. Uh well for the other ones that has a more of a general or specific pools uh That they definitely need. The mainline model is is the one that's more general in a sense. And that's sort of what Brian was referring to when he's talked about GPT-5's uh tools. Yeah, so the five five line non codex is more general across the board. It can restore respond to things that are it's it's much broader than just code it. It has coding capabilities that that are also mirrored in codecs and they they work together to keep that trued up. Since it's more general, it does have more sterability to different types of tools. And When you're implementing tools, you the model can get bogged down if it hasn't seen a a tool that it's used to and it might take more time thinking about how to use it or make more mistakes. So our recommendation is if you're wanting to go bleeding edge coding. Focused. Pay attention to the codex line and the codex STK and the codex models because that's the one that's like really aimed at that. You'll have to do you know, some work to like look at how we're implementing our tools inside of codecs to

Quote

Pause or not, I think isn't um as clear. This is what everyone's kind of pointing towards and there's a couple different reasons and a couple things that I think people should be doing uh if you were just laid off or if you're at a company that's doing these types of layoffs, there's a couple things that I will highlight that I think are actionable items. Okay. In addition to all of that, on the April 29th earnings. night alphabet shares rose six percent meta fell more than six percent microsoft and amazon both fell three percent we're gonna get into why and there's uh some pretty wild reasons First things first, let's talk about the AWSN OpenAI deal. What actually happened is that OpenAI is GPT 5.5 and also GPT 5.4, both of them are alive on bedrock. In addition to their codex platform and some of their other interesting things they have with their agent harness, all this is going to be put onto bedrock. Now the reason Why this is kind of big news because it's like, okay, great, it's like on another cloud platform. I think a lot of people don't understand that Microsoft Azure had an exclusive deal where OpenAI could only Only be on Microsoft Azure, uh because you know, Microsoft put ten billion dollars in and kicked off this whole, you know, wild AI feeding frenzy when Chat GPT first launched. And so open AI has been stuck over there. I actually think a lot of the problems th…

Quote

Um so next up we have uh Zvee. And I'm gonna pull him up. Yeah, good to see you. So Zvee is a uh prominent AI commentator. Um and he He writes a newsletter, the ZWE writes, on technical AI progress and Um he has been quite concerned about AI safety. In the last uh couple of weeks, um post Mythos, we have GPT 5.5. Uh Zvee, uh what are your initial uh reactions? So one thing I try to do is not jump to conclusions right away. So it's been less than twenty-four hours. We have G V D five point five and uh deep seek four. Uh over the last twenty four hours. So what I try to do is I try to let people try the model. I do all my queries with both the new model and everyone else's model. model at the same time and I read the model, yeah, I I say the I start to read the model. the model card, you know, and then I gather people's reactions and then I form a holistic judgment. And for me it's like it's too early, right? Like we booked this before we knew that was gonna be out it was like I I I don't want to jump to any conclusions, you know, Andon has had the model for a while they got to put it to a test and they got to see a bunch of results. They can draw like a lot more conclusions than I can.