28 November 2023

The Best & Worst GPTs + How To Make Your Own (#177)

2Frameworks

6Insights

Frameworks in this episode

Innovation5 steps

High-Value GPT Use-Case Filter

Prioritize GPT ideas with unique data, multimodality, or executable actions

Read →

Productivity5 steps

Recurring-Task GPT Training Loop

Turn frequent tasks into trained assistants through sustained iteration

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 1

Myth Buster15:00

Uploading Proprietary Data Does Not Keep It Private

The hosts warn that files uploaded to custom GPTs may be retrievable by users through carefully phrased requests. They connect this issue to the apparent removal of a GPT built around the Nomad List dataset, showing that a valuable data-backed assistant can also expose its core asset.

Uploaded GPT knowledge may be accessible to users
A user may be able to request the underlying JSON file
Proprietary datasets should not be uploaded without security testing
Data-backed GPTs create both value and disclosure risk

“the data you put into the gpts I don't know if you've seen this is accessible”
Kieran Flanagan · 15:30

“if you upload data and you don't want anyone else to see that data I can get from your GPT”
Kieran Flanagan · 15:30

#security#data-leakage#privacy#custom-gpts

Hot Take· 2

Hot Take03:30

Why Custom GPT Writing Still Feels Generic

Kieran tests a GPT trained on more than 50 of his LinkedIn posts and detailed formatting instructions. The result ignores the expected LinkedIn structure and adopts overwrought language, leading him to argue that writing remains one of AI's least impressive use cases.

Uploaded writing samples did not reliably reproduce the intended style
The GPT ignored explicit one- or two-sentence formatting guidance
Its prose sounded ornate and unnatural
Generic ChatGPT performed even worse in the comparison

“now off the bat you could see this isn't good because it's not even formatted for LinkedIn right”
Kieran Flanagan · 04:30

“I actually think writing is very underwhelming”
Kieran Flanagan · 05:30

#writing#linkedin#fine-tuning#content

Hot Take24:30

The GPT Store Could Drown in Low-Quality Tools

Kieran argues that the exceptionally low barrier to publishing custom GPTs will flood the store with marginal tools. Kip compares OpenAI's approach to Android's open marketplace, but notes that GPT creation has even less friction than conventional app development, increasing the burden on search and ranking.

Thousands of GPTs appeared shortly after launch
Minimal publishing friction encourages low-quality submissions
Users may default to ChatGPT instead of returning to custom GPTs
Discovery and ranking become critical in a noisy marketplace
Personal GPTs may be more useful than public-store GPTs in the short term

“the fact that there's no friction means that it's going to be overwhelmed with low quality just gpts with very little value”
Kieran Flanagan · 24:30

“I think in the short term it's going to be a lot of about us making gpts for us to use ourselves versus maybe going…”
Kip Bodnar · 26:30

#gpt-store#marketplaces#discovery#product-strategy

Story· 1

Story22:00

A GPT Narrates Webcam Movements Like a Nature Documentary

Kieran highlights a playful demonstration combining a webcam, GPT-4 Vision, and ElevenLabs. The system observes a person through the camera and narrates each movement in a David Attenborough-style voice, illustrating the unexpected experiences possible when multiple AI services are connected.

A webcam supplies live visual input
GPT-4 Vision interprets the person's movements
ElevenLabs generates the spoken narration
The demonstration combines perception and voice into a comedic experience

“he hooked up G uh his webcam to gbt for vision and then he used 11 labs to have a David ador's voice narrate his…”
Kieran Flanagan · 22:30

#vision#voice-ai#elevenlabs#demo

Tool· 1

Tool23:00

Two Ways to Find Useful GPTs Before the Store Works

The hosts present two discovery tactics: use a recommendation GPT that maps a described use case to another GPT, or search Google's index for GPT pages on OpenAI's domain. Adding keywords such as marketing, design, stocks, or sales narrows the results.

A recommendation GPT can surface tools for a described use case
Google can search indexed GPT pages on OpenAI's site
Topic keywords narrow the available GPT results
Thousands of GPTs were already discoverable

“you basically use this GPT to explain your use case and it will then surface the best GPT you should go and use to solve…”
Kip Bodnar · 23:00

“if you do site colon chat. open.com slgg you start to see I I I don't know this cover there's about 6,970 here”
Kieran Flanagan · 23:30

#search#discovery#gpt-store#google

Takeaway· 1

Takeaway19:00

Advanced Users May Prefer Configuration Screens to AI Chat

Kieran compares creating a GPT conversationally with editing its configuration directly. Although natural language lowers the entry barrier, repeated back-and-forth can be slower than explicit controls for experienced users who already understand what they want to configure.

Natural-language setup helps users begin quickly
Direct configuration can be faster for precise changes
Advanced users may prefer explicit controls
Future software may need both conversational and conventional interfaces

“the more I created Bots the more I found myself actually going into the configure screen”
Kieran Flanagan · 19:30

“there really is a place still for the click click click set things up yourself”
Kieran Flanagan · 20:00

#ux#configuration#natural-language#software-design