Custom GPTs Just Leveled Up But Are They Breaking? (Ep. 485)

Want to keep the conversation going?Join our Slack community at thedailyaishowcommunity.comThe team runs a grab bag of AI updates, tangents, and discussions. They cover new custom GPT model controls, video generation trends, Midjourney’s 3D worldview, ChatGPT's project features, and Apple's recent AI research papers. The show moves fast with insights on LLM unpredictability, developer frustrations, creative video uses, and future platform needs.Key Points DiscussedCustom GPTs can now support model switching, letting both builders and users choose the model best suited for each task.Personalization and memory features make LLM results more variable and harder to standardize across users.Clear communication and upfront expectations are essential when deploying GPTs for client teams.Midjourney is testing a video model with a 3D worldview approach that allows for smoother transformations like zooms and spins.Historical figure vlogs like George Washington unboxings are going viral, raising new concerns about AI video realism and misinformation.Credits for video generation are expensive, especially with multi-shot sequences that burn through limits fast.Custom GPT chaining may be temporarily broken for some users, highlighting a need for more stability in advanced features.ChatGPT Projects received updates like memory support, voice mode, deep research tools, and better document sharing.Despite upgrades, projects still do not allow including custom GPTs, limiting utility for advanced workflows.Connectors to tools like Google Drive, Dropbox, and CRMs are becoming more powerful and are key for real enterprise use.Consultants need to design AI solutions with the future in mind, anticipating automation and agent orchestration.Apple’s recent papers were misinterpreted. They explored limitations in logical reasoning, not claiming LLMs are fundamentally flawed.Timestamps & Topics00:00:00 🧠 Intro and grab bag kickoff00:01:27 🛠️ Custom GPTs now support model switching00:04:01 🔄 Variability and unpredictability in user experience00:06:41 💬 Client communication challenges with LLMs00:10:11 🪴 LLMs are more grown than coded00:13:51 🧪 Old prompt stacks break with new model defaults00:16:28 📉 Evaluation complexity as personalization grows00:17:40 🧰 Custom GPT apps vs GPTs00:19:22 🚫 Missing GPT chaining feature for some users00:22:14 🎞️ Midjourney video model and worldview00:27:58 🎥 Rating Midjourney videos to train models00:30:21 📹 Historical figure vlogs go viral00:32:38 💸 Video generation cost and credit burn00:35:32 🕵️ Tells for detecting AI-generated video00:38:02 🗃️ ChatGPT Projects updates and gaps00:40:07 🔗 New connectors and CRM integration00:43:40 🤖 AI agents anticipating sales issues00:46:26 📈 Plan for AI capabilities that are coming00:46:59 📜 Apple research papers on LLM logic limits00:51:43 🔍 Nuanced view on AI architecture and study interpretation00:54:22 🧠 AI literacy and separating hype from science00:56:08 📣 Reminder to join live and support the show00:58:21 🌀 Google Labs hurricane prediction teaser#CustomGPT #LLMVariance #MidjourneyVideo #AIWorkflows #ChatGPTProjects #AgentOrchestration #VideoAI #AppleAI #AIResearch #AIEthics #DailyAIShow #AIConsulting #FutureOfAI #GenAI #MisinformationAIThe Daily AI Show Co-Hosts: Andy Halliday, Beth Lyons, Brian Maucere, Eran Malloch, Jyunmi Hatcher, and Karl Yeh

There are no comments yet.

Please log in to write the first comment.