GPT-5 Arrives, and We Try the New Alexa+  ## Metadata - Author: [[Hard Fork]] - Full Title: GPT-5 Arrives, and We Try the New Alexa+ - Category: #podcasts - URL: https://share.snipd.com/episode/d72bafc1-0c13-4510-a291-706e473646c1 ## Document Notes ## Highlights - **Smelling Salts And Fainting** - Kevin Roose fainted during a blood draw and was revived with smelling salts. - The startling revival felt surreal and intensely physical. Transcript: Speaker 4 Over the last few decades, the world has witnessed incredible progress. From dial-up modems to 5G connectivity. From massive PC towers to AI-enabled microchips. Innovators are rethinking possibilities every day. Through it all, Invesco QQQ ETF has provided investors access to the world of innovation with a single investment. Invesco QQQ, let's rethink possibility. There are risks when investing in ETFs, including possible loss of money. ETFs' risks are similar to those of stocks. Investments in the tech sector are subject to greater risk and more volatility than more diversified investments. Before investing, carefully reading Consider Fund Investment Objectives, Risks, Charges, Expenses, and More in Perspectives at Invesco.com. Invesco Distributors Incorporated. Kevin Roose I had a first this week. What's that? I had my first experience with smelling salts. Wait, did you faint? Yes. So I had to get a blood draw at the doctor, and I am a big baby when it comes to getting blood taken. And so like half the time when I get blood taken, I pass out. And this time, I not only passed out, but I vomited. And had to be brought back with smelling salts. ([Time 0:00:00](https://share.snipd.com/snip/23fd387b-ceb0-45e0-a1ed-e4346aefe330)) - **Automatic Model Router** - OpenAI built a router to analyze requests and route them to the right model. - This hides model selection from users and applies reasoning models when needed. Transcript: Kevin Roose So one of the most annoying features of ChatGPT for years now has been this model selector where you go in and it sort of gives you this little drop-down menu and it defaults to GPT-4 now, But you can kind of pick your own model if you want something more powerful than that and you're a paying user. But for GPT-5, OpenAI is doing away with the model picker or at least making it sort of less necessary because they have built this, what they called a router that will essentially analyze Your request and how much sort of computation it needs to be able to answer that request and whether it's a simple query or something more involved and it will kind of direct it to the correct Model. And I think for a lot of people, this is going to be their first experience with a reasoning model because OpenAI does not make that the default right now in ChatGPT. And so I think that will be a ([Time 0:07:31](https://share.snipd.com/snip/e717c7d0-1b41-4301-8665-41b1ab32984c)) - **Extensive Red‑Teaming Revealed** - OpenAI performed 5,000 hours of red‑teaming and consulted external experts for GPT‑5. - They rated the model as high risk for enabling novel biological misuse and added protections. Transcript: Casey Newton Come out is what was the safety testing experience like? And is this model going to be sycophantic? And what sorts of very intense relationships are people going to form with it? And Nick Turley addressed that one. He noted that earlier this week, OpenAI put out a blog post, which actually wrote about in Platformer this week, that is all about their approach here. They say they're working with physicians, really trying to bring in a lot of outside expertise to help them understand how people are interacting with these models and make them safer. And he said that they are absolutely not optimizing for engagement here. They just want to sort of make a useful tool that sends you on your way and that essentially they're going to have more to communicate about this soon. So we didn't get a ton of detail there, but they have said that at least in some ways they think that they have improved these models to make them less sycophantic. In addition to that, they said they did 5,000 hours of red teaming. They shared, I believe, these models with some external experts for advice on that. So they did say that they rate this model as high on the scale of could it be used to create novel bio risks. So they're building in a bunch of protections around that. But that doesn't seem great. But anyway, that was the sort of safety report that we got in advance of the launch. ([Time 0:13:20](https://share.snipd.com/snip/0481a114-b8f3-4f7c-997c-6b717d3fd52d)) - **GPT‑5 Raised The Floor** - GPT‑5 didn't dramatically raise the AI frontier's ceiling but raised the floor for users. - Free users now access stronger reasoning that makes advanced capabilities common. Transcript: Kevin Roose Right, Casey, it is now Thursday. GPT-5 has been officially released for a few hours. I still do not have access to it for some reason, but I gather that you do. So give me your day one vibe check. What are you seeing? What is GPT-5 like? And what do you make of the reaction to it? Casey Newton Well, this is a very significant moment in the history of Hard Fork, Kevin, because for the first time, I'm having a conversation with you while vibe coding something. What are you vibe coding? Well, ChatGPT is currently hard at work building a to-do list app for me. I said I wanted it to have the aesthetic of the Fantastic Four First Steps movie that just came out. I didn't love the movie, but I did love the production design. So I was like, make me a to-do list app that looks like that. Let's see how it goes. Kevin Roose I can't believe we're building giant gigawatt data centers for your stupid to-do apps. This is so wasteful. God. Casey Newton Listen, you need to send over Roosbench, your proprietary suite of evals, so I could really put this thing through its paces. But look, let me give you some high-level notes, Kevin, on what I'm seeing and on what others are seeing. The headline here is that this does seem like a really meaningful improvement to ChatGPT. And I think in particular, if you are a free user of ChatGPT, you're going to have a great day, right? Because for the first time now, in addition to the kind of, you know, standard ChatGPT model, you're going to have some reasoning capabilities. So essentially, if you're cheating your way through high school, you're just going to have a lot easier time of it now because this thing can do some really extended work on long problems. Kevin Roose Yes, you can now cheat your way through an entire semester with just one press of a button. Exactly. This is good. Yeah, I mean, the way I saw people talking about it online was that they thought that OpenAI had not raised the ceiling of the AI frontier by a lot with GPT-5, but they had raised the floor. Essentially, now all the free users who ([Time 0:18:39](https://share.snipd.com/snip/a60d7b50-bac2-410d-a952-bbd6c0b12e64)) - **Double-Check Mission‑Critical Output** - Don't trust GPT‑5 for mission‑critical facts or decisions. - Always verify model claims with authoritative sources before acting. Transcript: Kevin Roose And one of the things they showed was that hallucinations, the rate of GPT-5 just making stuff up while answering questions, has gone way down. It's now, you know, for some types of questions, it's sort of around 1% hallucination rate. And I think that was interesting to me because this is clearly something that was a problem with earlier versions of this. In fact, there was some speculation and some indication that actually newer, these newer reasoning models were hallucinating at higher rates than the sort of previous generation Of models. And there was a lot of concern about that. And it seems like they have figured out a way to get the hallucinations under control with GPT-5. Although with everything, I don't totally trust these benchmarks. I'm going to have to see this for myself. Casey Newton Yeah, everyone's mileage is going to vary on this one. I will say I have already caught it hallucinating a couple of times, like somewhat disappointingly. So as always, don't trust these things for anything mission critical. You're always going to want to double check your facts. ([Time 0:23:53](https://share.snipd.com/snip/9e6be0cc-a34e-4b74-a6f9-e1874fc79d8d)) - **Echo Show Ads Drove User To Unplug** - Casey Newton unplugged an Echo Show after onboarding repeatedly pushed Amazon product ads. - Persistent purchase prompts ruined the device's initial appeal for her. Transcript: Casey Newton And so the thing shows up at my house. And basically what I've come to understand is that an Echo Show is a device that just constantly invites you to spend money with Amazon. And I found it honestly infuriating because I plugged this thing in and when you set it up, it's like, what, you know, what kind of background do you want? I was like, show me some art, you know, that's one of the options. And I would say for about four seconds per minute, it would show me, you know, some Renaissance, you know, masterpiece or something. And then it would be like, Hey, do you want aspirin? You want paper towels? You want to buy paper towels? You can actually buy paper towels right now. Just say, hey, Alexa, buy paper towels. And it was just sort of this like forever. And so I eventually just unplugged the thing because I was like, why did I just spend $90 to have a permanent rotating advertisement for household products like on my desk? That is so weird. So I was like, it just put such a bad taste in my mouth about the whole thing. ([Time 0:36:31](https://share.snipd.com/snip/ebab7441-707c-49e2-a212-e57040e45931)) - **Core Feature Regression: Alarms** - Kevin Roose tried to cancel a scheduled alarm and Alexa Plus ignored the command, then rang anyway. - Regressions in core routines erode trust faster than new features build it. Transcript: Kevin Roose Wait, so tell me how it got it wrong. So the new Alexa Plus update seems to have broken Alexa's ability to reliably set and cancel which is a core thing that I use this product for. And so, for example, this morning, I woke up on my own a little bit earlier than my alarm, like 10 minutes before my alarm was supposed to go off. And so I said to Alexa, Alexa, cancel the alarm. Silence. Nothing. This is a command that I have issued probably a thousand times. And Alexa Plus is a little smarter now, and she's giving you the cold shoulder. Yes, and she's saying, actually, I'm going to wake you up anyway in 10 minutes. So that was not good. ([Time 0:40:56](https://share.snipd.com/snip/ce3fa393-b910-4afe-b197-3b50235852d8)) - **Prefer Editable Chat For Complex Tasks** - Ask complex or research tasks in a chat interface where you can edit outputs. - Avoid relying on voice assistants for answers you must refine or verify. Transcript: Casey Newton That when I'm using a chat bot on my laptop and it gives me something that's like 80 or 85% right, that's much more useful to me than like an Alexa response that's 85% right. Because on a chat, like in a chatbot setting, I can just sort of take what I need. I can edit or modify it. I can maybe ask the same question of another chatbot and see if I get a slightly different or better result. Like I feel much more in control of my own destiny. I can take the stuff that works and leave behind the stuff that doesn't. When you're doing this thing with a smart speaker, if it doesn't work, you say, yeah, why do you spend 90 bucks on this piece of junk? ([Time 0:46:50](https://share.snipd.com/snip/94d5d19e-d2f9-40cb-b18c-31749ffef7c6)) - **LLM And API Translation Problem** - LLMs speak natural language while APIs require structured commands, creating an interface gap. - Amazon built orchestration layers to translate intents into predictable API calls. Transcript: Daniel Rausch System. I get it. So, you know, we want to capture all the benefits of that non-deterministic. We'd call it stochastic system in the space. It has the elegance of really engaging in human conversation, but we want the predictable outcomes. Now, large language models don't support interfaces out of the box to classic systems. So getting those capabilities to interface, we would talk about it as APIs across these interfaces for other systems. It's quite hard. They speak natural language. APIs don't speak natural language. They speak clunky computer science language, but it's very predictable and it gets a lot of things done. ([Time 0:51:32](https://share.snipd.com/snip/4e362ad0-231d-44e8-b7a9-b1cda348c18b)) - **Alexa Uses Many Specialized Models** - Alexa Plus orchestrates over 70 specialized models to handle varied tasks. - Most inference traffic runs through Amazon Nova models for fine‑tuned control. Transcript: Daniel Rausch I would love it if it were that easy. You need a set of models. There are over 70 models in Alexa Plus. It's a vast space. There are different models specialized in different tasks. There are different corpuses of training data we use on different models to get them to complete instruction sets for us and really follow the rules of the road in interfacing with something. You always need to loop back to central systems that are maintaining context in the conversation and picking up from references and pronouns that you've used to refer back in time, Right, and sort of cascade those forward. But the amount of work that went into just the interface between a large language model and the downstream systems that complete tasks is, I mean, it's the biggest body of work that we've Put in. And there's, it would be too much to even try, without a whiteboard here, it would be too much to even try to explain to you and your listeners, I think, the technical depth that went into It. We've got a great team working on it and it's hard. Kevin Roose Of those 70 models in Alexa Plus, how many are Amazon's own in-house models versus models like Claude that you all get from external companies? There's a mix. Daniel Rausch So, you know, the best way to know what models are in Alexa Plus is just go to the Bedrock webpage and look at an update on the latest in there. We use the best tools that we have available to us for the job and we've got great partners over in AWS helping make sure we've got the right best tools for the job. Most of our traffic does flow through Amazon Nova models. We have the most control over how those get trained and tuned and post-trained. I think it's over 80% of traffic on sort of the main big inferences within the system flow through Nova models. ([Time 0:58:21](https://share.snipd.com/snip/820c4f13-48ec-4dcc-aafb-6cfd87a2cb88)) - **Opt Out If Alexa Frustrates You** - If Alexa Plus frustrates you, opt out; Amazon made switching back easy. - Try early access then revert until features meet your needs. Transcript: Daniel Rausch But I, you know, I will tell you that the body of feedback that we get from customers doesn't accord with that specific version of it. Definitely customers want to learn what they can do. That is one of the biggest things that we hear from customers. I want to come back to what you said about, you know, unplugging the device and plugging it back in. We made the Alexa Plus experience incredibly easy to get out of and get back into and get out of, which is not true for sort of like an OS update, right? It's very hard to go backwards. And we worked very hard to try to make it possible because we knew that it would be so much change. The very high 90 percentile of customers stick to the new experience. ([Time 1:08:00](https://share.snipd.com/snip/9a3c12a3-2051-4908-aa9e-8111167a4940)) - **Prime Integration Over Ads** - Amazon embedded Alexa Plus in Prime to boost value and cross‑service engagement. - Advertising is part of the plan but Prime integration is the main business lever. Transcript: Daniel Rausch I definitely don't think you'll inevitably be more annoyed. I would say advertising is definitely part of the business plan, but it's not the biggest part. It's actually probably the smallest part. The most important decision we made on the business side with Alexa Plus was bringing it into Prime. Both all of a customer's prime benefits, because you might watch a video or listen to a song from Amazon Music or use your Amazon Photos benefit, which is awesome with an Echo Show and Review your family photos. I use that all the time to look back at the kids in particular. But, you know, you have this long list of prime benefits. Alexa is a great place where they come together and putting in the value of having this world's best personal assistant into Prime just turns the Prime flywheel. And we know every time we've added a benefit to Prime, customers use their Prime benefits more. It's stickier to them. It provides them more value. And it turns into a great business. ([Time 1:10:00](https://share.snipd.com/snip/ae030d03-894c-4486-a039-ebf5e19c3ddd))