This Week's Releases from OpenAI | EP 148

Quick Overview

OpenAI released GPT-5, which Sam Altman described as a major upgrade and a significant step towards AGI, comparing it to talking to a "PhD in a subject" versus a "high school student" for GPT-3 and a "college student" for GPT-4, though it does not continuously learn. Concurrently, Amazon launched an overhauled Alexa Plus, integrating generative AI, which offers a more conversational and personalized experience but suffers from latency, broken basic functions like alarms, and persistent upselling, leading to a mixed reception and highlighting the challenges of integrating new AI into legacy systems.

Key Points: OpenAI released GPT-5, which CEO Sam Altman claims is a major upgrade and a significant step towards AGI, stating, "GPT5 he said is uh the first time it feels like talking to an expert, someone you know who has a PhD in a subject." Altman clarified that GPT-5 does not continuously learn, which he considers a prerequisite for AGI, and that while previous models felt like talking to a "high school student" (GPT-3) or "college student" (GPT-4), GPT-5 feels like conversing with an expert. Amazon launched Alexa Plus, an AI-powered overhaul of its virtual assistant, offering more conversational capabilities and integrations, but early users reported issues like latency, broken alarm functions, and persistent product advertising. One host described the Alexa Plus experience as infuriating, stating, "I was just like, "Why did I just spend $90 to have a permanent rotating advertisement for household products like on my desk? That is so weird." Despite improvements in voice and conversational flow, Alexa Plus was noted to regress on basic functions, with one host experiencing, "the new Alexa Plus update seems to have broken Alexa's ability to reliably uh set and cancel alarms, which is a core thing that I use this product for." The hosts concluded that integrating advanced AI like LLMs into legacy systems is a significant challenge, leading to a "messy fit" and unreliability, with one host stating, "The finest minds in the world who are working on this stuff actually don't know how to do this yet." OpenAI also released two open-source models, aiming to compete with other AI providers, and claims GPT-5 has a reduced hallucination rate of around 1%, though hosts remain cautious and plan further testing.