Exploring Ear Training with a Pitch Identification Game

The team tees up a show focused on real GPT 5 use cases. They set expectations after a bumpy rollout, then plan to demo what works today, what breaks, and how to adapt your workflow.

Key Points Discussed

• GPT 5 launch notes, model switcher confusion, and usage limits. Plus users reportedly get 3,000 thinking interactions each week.

• Early hands on coding with GPT 5 inside Lovable looked strong, then regressed. Gemini 2.5 Pro often served as the safety net to review plans before running code.

• Sessions in code interpreter expire quickly, which can force repeat runs. This wastes tokens and time if you do not download artifacts immediately.

• GPT 5 responds best to large, structured prompts. The group leans back into prompt engineering and shows a prompt optimizer to upgrade inputs before running big tasks.

• Demos include a one shot HTML Chicken Invaders style game and an ear training app for pitch recognition, both downloadable as simple HTML files.

• Connectors shine. Using SharePoint and Drive connectors, GPT 5 can compare PDFs against large CSVs and cut reconciliation from hours per week to minutes.

• Data posture matters. Teams accounts in ChatGPT help with governance. Claude’s MCP offers flexibility for power users, but risk tolerance and industry type should guide choices.

• For deeper app work, consider moving from Lovable to an IDE like Cursor or Cloud Code. You get better control, planning, and speed with agent assist inside the editor.

• Gemini Advanced stores outputs to Drive, which helps with file persistence. That can outperform short lived code interpreter sessions for some workflows.

• Big takeaway. Match the tool to the task, write explicit prompts, and keep a second model handy to audit plans before you execute.

Timestamps & Topics

00:00:00 🎙️ Cold open and narrative intro

02:18 🗓️ Show setup and date, who is on the panel

02:43 🧭 Today’s theme, GPT 5 use cases and rollout recap

05:39 🧑‍💻 Lovable coding with GPT 5, early promise and failures

07:44 🧪 Switching to Gemini 2.5 Pro as a plan validator

09:55 ❓ GPT 5 selection disappears in Lovable, support questions

10:08 🔁 Hand off to panel, shared issues and lessons

10:08 to 13:38 🧵 Why conversational back and forth stalls, need for structure

13:38 ⏳ Code interpreter sessions expiring quickly

15:00 🧱 Prompt discipline and optimizer tools

16:54 💸 Theory on routing and cost control, impact on power users

19:45 🔀 Model switcher has history, why expectations diverge

20:48 👥 GPT for mass users versus needs of power users

23:19 ⚙️ Legacy models toggle and model choice for advanced work

25:04 🧩 Following OpenAI’s prompting guide improves results

27:10 🔧 Prompt optimizer walkthrough

29:31 🐔 Game demo, one shot HTML build and light refinements

31:13 💾 Persistence of generated apps and downloads

32:42 🔗 Connectors demo, PDFs versus CSVs at scale

34:58 ⏱️ Time savings, hours down to minutes with automation

36:43 🛡️ Data security, ChatGPT Teams, and governance

39:49 🚫 Clarifying not Microsoft Teams, Claude MCP option

41:20 🗺️ Taxonomy visualizer and chat history exploration

45:36 📉 CSV output gaps and reality checks on claims

47:30 🧭 UI sketch for a better explorer, modes and navigation

48:47 🛠️ Advice to move to Cursor or Cloud Code for control

52:49 📚 Learning path suggestion for non engineers

55:42 🎼 Ear training app demo and levels

59:07 🔄 Gemini versus GPT 5 for coding and persistence

60:30 🗂️ Gemini Advanced saves files to Drive automatically

63:06 🧳 Storage tiers, Notebook LM, and bundled benefits

64:18 🌺 Closing, weekend plans, and community invite

The Daily AI Show Co-Hosts: Andy Halliday, Beth Lyons, Brian Maucere, Eran Malloch, Jyunmi Hatcher, and Karl Yeh