AI Enters a New Era With GPT-5  ## Metadata - Author: [[Hard Fork AI]] - Full Title: AI Enters a New Era With GPT-5 - Category: #podcasts - URL: https://share.snipd.com/episode/3259f2f9-00ad-4312-8eee-f40732839a76 ## Document Notes ## Highlights - **AIbox Beta Lets You Test Multiple Models** - Jaeden Schafer says AIbox.ai is a beta no-code app builder giving access to 40 top AI models in one place. - He invites listeners to test and switch models mid-chat without multiple subscriptions. Transcript: Jaeden Schafer Before we do, I wanted to mention, if you want to try all of the models I talk about on the podcast, I'd love for you to try out my own software company called AIbox.ai. We're currently in beta. We're building this super cool drag and drop no-code app builder, which is basically a competitor to Google Opal, but it's got all of the AI models on it. But currently today, so this is not accessible. This is an exciting sneak peek that we're building out right now. Currently, what you have is access to the top 40 different AI models all in one place for one price. So you can chat with it just like you would chat GPT, but you can switch to any different model that you want to try midway through. So maybe you want to try out the latest model from DeepSeek and compare it to something from Anthropic. It's all on there. It's a great platform. Link is in the description. It is AIbox.ai. ([Time 0:00:55](https://share.snipd.com/snip/65bb0022-5296-46a5-9dc9-83bfcd3d27b8)) - **Three New Models Replace Old GPT Lineup** - OpenAI consolidated older GPT variants into three models: GPT-5, GPT-5 mini, and GPT-5 nano. - The system will auto-select the best model for queries so users needn't pick manually. Transcript: Jaeden Schafer One thing I'm very excited about is currently, if you go to chat GPT, and you click the drop down, it's going to ask you, do you want to use GPT 40, 03 04 mini 04 high, and a whole bunch of other Ones. All of those models are getting depreciated, disappeared, gated, and instead it's going to be replaced by three new models, GPT-5, GPT-5 mini, and GPT-5 nano. So those are the three models. The nano is like the super tiny version that's going to be able to run on edge devices. Mini is just a faster, quicker, less powerful. And then GPT-5 is their best model. And the great thing is you don't have to pick which model you want to answer your questions. You don't have to pick what tool, right? Because sometimes we're looking at things like, you know, 4.0 Mini High, great at coding and visual reading. 0.3 uses advanced reasoning, right? Like you're like, do you want to use the reasoning? There's all these different tools that you could toggle on and off if you want to do deep research and agent mode and all this kind of stuff. Okay. You don't have to, other than I guess agent mode probably, but you don't have to ask it or pick your model anymore. You just ask a question and it will choose what the best model is for you. ([Time 0:01:56](https://share.snipd.com/snip/2c2798e7-97cd-4c74-aa63-e261d3ba7ec1)) - **Fewer Hallucinations, Marginal Accuracy Gains** - Jaeden Schafer says GPT-5 is more accurate and hallucinates far less than prior models. - The gains are real but marginal, and some models still beat GPT-5 on benchmarks. Transcript: Jaeden Schafer Number one, improved answer quality. Basically, it's more accurate. It makes way fewer mistakes. It hallucinates way less. So that's obviously going to be pretty exciting, just that it's marginally better. How much better? We'll look at some benchmarks. Some people were kind of roasting it. It made improvements, but it wasn't as huge. Everyone was expecting it to make this huge step game where it was going to be insanely better and absolutely crush everything. And the hype was really high um at the end of the day i think it is better but not insanely better and there are some models that on different benchmarks are still beating it which is not Really a good place to be you you probably want an opening eye to beat everybody before they put out an update especially because we have google cooking up something, but more on that In a minute. ([Time 0:03:21](https://share.snipd.com/snip/fd93eb10-894a-4046-950c-cf90474e0a0d)) - **Huge Context Window Improves Long-Term Memory** - GPT-5 expands context capacity toward a one million token window to recall long conversations. - This reduces forgetting in long threads and improves continuity across projects. Transcript: Jaeden Schafer So expanded reasoning and context. One thing I am very excited about, Google's been doing this for a long time, but they just added 1 million token context window. Now this is something that Gemini has been doing, but OpenAI was not. So basically it's going to be able to look at all of your past conversations that you've had with it. And it's going to be able to look at like way further back. We kind of had this problem with ChaiGPT where you'd like, if you had a thread open for a very long time and start forgetting things that you said at the beginning of the thread, this probably Is going to be less of an issue now as the context window is way, way bigger and it can remember way more things. ([Time 0:04:05](https://share.snipd.com/snip/3671f850-d4e0-4c73-81d0-c704b76f08ce)) - **Stronger Coding And Task Automation** - Jaeden Schafer notes GPT-5 improves coding speed and iteration on benchmarks. - Its better multi-step task automation makes agents more capable at complex workflows. Transcript: Jaeden Schafer One other thing that's interesting is the coding and task automation has improved. So coding basically is just doing better on the benchmarks at coding tasks, not just writing the code, but really quickly iterating on the code. It's getting faster at that. This is something that Claude has been crushing and winning in the code, like hands down. Claude is a preferred coding tool for developers, but it seems like OpenAI is trying to put a really big effort in here and try to basically keep up. They have a whole bunch of demos that you can basically make, you can use and create all sorts of things. They have in their blog post this like, it's called jumping ball runner. And the whole thing is that whole game where you click a button and the ball is jumping over the things. The whole thing was just created in GPT-5. You put a prompt in, create an entire game, the color, the design, all of that. Very, very interesting. One other thing that's pretty great is the task automation. So basically it's getting better at multi-step prompts. If you tell it like, hey, you know, create a newsletter for me, right? Or I don't know, something like that. And then it will basically break that into down into multiple steps. Or if you're like, create a newsletter and also write a blog post and also write a link or like a post on LinkedIn based off of this data that I give you, right? It's much better doing these multi-step tasks, which is like sure good for like chat GPT. But really what it's getting at is it's going to be much better for agents. ([Time 0:04:38](https://share.snipd.com/snip/b876d06c-aac8-4d79-af79-a90035e8c826)) - **Multimodal Inputs: Video, Audio, Images** - GPT-5 accepts images, audio, and video and links them to OpenAI's other tools like Whisper and Sora. - ChatGPT now consolidates transcription, media understanding, and generation into a single interface. Transcript: Jaeden Schafer So that's interesting. The modal input and output is another thing that I am absolutely excited about. Basically what that means is you're going to be able to upload images and or audio and video as inputs into the model. So you upload a video and it's going to understand. Now I think this is really important, especially as OpenAI starts tying in everything they're doing with OpenAI's Sora, their video generator. They have a video generation tool. They have an audio generation tool. You also are going to be able to get, you know, transcripts from videos that you put in and transcripts from audio that you upload. Now, these are all tools that OpenAI has tools for. They have Whisper, which helps you to do that basically. But now it's everything's just getting tied into ChatGPT. So all of these tools that they had that some of them were disconnected, it feels like we're consolidating everything, even to the point where they took their, you know, their agent, And they put the agent into chat GPT, it feels like everything's coming into chat GPT, which makes it way more useful of a tool. ([Time 0:06:04](https://share.snipd.com/snip/bac44f05-cb12-4dd0-b521-f9b36456ddf3)) - **Longer-Term Memory With User Controls** - GPT-5 includes longer-term memory for paid users that spans months and many projects. - Users can review, edit, and delete stored memories to control what the model remembers. Transcript: Jaeden Schafer Okay. One other thing that it's getting a lot better at is basically keeping track of your recent conversations. It's the memory feature for paid users. It now includes a longer term memory and it's able to go across a whole bunch of months and like many months and many projects that it's able to remember so basically the scope of its memory Is increasing they're basically just giving you like more memory storage on their servers which i honestly i think that's great um also like before users can review edit and delete What the model remembers. So if you don't want it to remember a specific thing, you can go see what it's remembering about you and go delete that out of its, out of its memory. Uh, honestly, this happens to me sometimes where like, I'll be, I'll be helping somebody else on a project and they're like, Hey, like my mechanic shop needs to do X, Y, and Z with AI. Can you help me test this thing out? And I'm like, yeah, sure. Like, here's how I do it. And I go on there for like a client or something and whip something up. And then before you know it, I all of a sudden have like embedded in the chat to memory. It's like, you own like Gary's Auto Shop. Like, do you want to get some great marketing tips for this? I'm like, oh no. So that kind of stuff you want to be able to pull out. ([Time 0:08:16](https://share.snipd.com/snip/05beeda2-6937-4eb5-8bcc-ddc1bd51ca69)) - **Market Reaction: Google Surges After Launch** - Polymarket bets flipped after GPT-5's release, dropping OpenAI from 73% to 13% and lifting Google to ~82%. - Traders now see Google as likely to release a superior model soon. Transcript: Jaeden Schafer It's got a bunch of cool features. So where is it? How is it being received? I would say I made a bunch of posts over on LinkedIn, but one of them that was interesting, I went over and took a screenshot of Polymarket. Now, if you look at Polymarket, basically, there's a bet on there with $2 million being bet right now that says which company has the best AI model by the end of August, right? And I mean, it's only the 7th of August right now. So before today, all of the betting was basically saying, I think as of like this bet has been going on, I guess the bet kind of opened up on July 1st. So over a month ago, this bet opened up. OpenAI took the lead on the bet as of June 26th. So everyone knew that GPT-5 was coming out. They took the lead to the point where OpenAI was at like 73% favored to win this bet. 73% of people said OpenAI is going to have the best model by the end of August. They dropped their model on the 7th. And instantly, as soon as they drop it, OpenAI crashes to 13%, and Google spikes up to 82%. That is a massive reversal. Basically, everyone was really hyped that GPT-5 was going to blow everyone out of the water, and it didn't. It was marginally better than some models. ([Time 0:09:25](https://share.snipd.com/snip/8284e982-0f61-4246-9c8c-d185068d3fd1)) - **Hype Backlash: AGI Is 'Canceled' Narrative** - Social media voiced disappointment, declaring 'AGI is canceled' after GPT-5 underwhelmed. - Jaeden Schafer argues development will continue rapidly despite the backlash. Transcript: Jaeden Schafer The first one being that it seems like we have really this, it feels like a lot of people are saying AGI is canceled. If you go over on Twitter, everyone's saying AGI is canceled. It's no longer going to be, you know, we have time to make all of our, make our money before the AI takes over the world and it's kind of sad because on the one hand like GPT-5 was supposed It was so hyped and it was going to be the biggest thing of all time and when it didn't quite meet people's expectations a lot of people are like oh man it's going to be 30 years before I ever See anything cool is this reality in my opinion it's not in my opinion I think we're going to see all sorts of really exciting updates and advancements that AI was going to keep moving Quite rapidly we see a lot but i think that at the end of the day um this probably was a little bit disappointing because there was so much hype around gpt5 i mean we saw the length that opening I went to i remember when gpt4 came out everyone was blown away then gpt5 it was a huge jump in quality. And then, or GPT-4 And then opening, I did literally everything humanly possible to not release GPT-5. They released GPT-4 and O3 and O3 Mini and Nano and GPT-4 research, like all of these random naming conventions, basically, because I didn't want to release GPT-5 and it be a letdown. So it felt like to a lot of people, by the time they released GPT-5, it was going to be this big, huge, basically, moment where everyone was, I don't know, super stoked about what actually Came out. ([Time 0:12:28](https://share.snipd.com/snip/19ff131a-663c-49da-9615-a5d25bdfc8c3))