ChatGPT & OpenAI
The library's default general-purpose LLM, used as a research analyst, prompt-builder, travel planner, and image-prompt generator — alongside a running skeptic's counterpoint.
Across these highlights, ChatGPT is the general-purpose LLM the collection keeps returning to — less a single tool than a Swiss-army knife pointed at whatever task is at hand: reading a 300-page S-1 and writing an investment thesis, planning travel, building Midjourney prompts, wrapping itself in a Telegram bot, or mining a startup idea. The flagged passages trace an arc from December 2022 (early ChatGPT tinkering) through Deep Research and Sora in 2025. Running alongside the enthusiasm is a deliberate counterweight — the skeptic's case that "AI" is really just automation — which the collection saves in the same breath as the playbooks.
The default general-purpose tool
What makes ChatGPT the default here is breadth. The same product shows up doing wildly different jobs, and the highlights capture each one as it was discovered:
| Use case | What was highlighted | Source |
|---|---|---|
| Financial research | Deep Research reads DoorDash's S-1 + 10-Ks, writes an investment thesis1 | @buccocapital |
| Travel planning | "The real killer use case for ChatGPT in travel"2 | @gshewakr |
| Miles & points | Tested on airline award routing — and found lacking3 | Aaron Wong |
| Image prompting | Asked to write a better Midjourney prompt for an app icon4 | @_nghiatran |
| Software generation | Specialized prompts vs. default: "night and day"5 | @mckaywrigley |
| No-code building | A lo-fi beats radio player in 10 steps, 90 minutes6 | @charlierward |
| Tooling / access | Wrapped in a Telegram bot for easy daily access7 | @m1guelpf |
| Idea generation | Mine your own ChatGPT history for repeated prompts8 | @gregisenberg |
The earliest highlights — Miguel Piedrafita's Telegram wrapper and Nghia Tran's Midjourney experiment, both flagged 29 December 2022, weeks after launch — are about simply getting to the thing more conveniently. "Got tired of opening the GPTChat website every time, so I made a Telegram bot" captures the moment before ChatGPT had an app, a plugin store, or Deep Research.7
ChatGPT as a research analyst
The most heavily highlighted source is @buccocapital's walkthrough of using Deep Research to analyze DoorDash — and the reason the whole thread was flagged is the reusable playbook buried in it. The clever move is recursive: "I asked ChatGPT to build me a prompt for Deep Research to do Deep Research on Deep Research prompting."1 ChatGPT reads the literature on good prompting, then emits a prompt template that "routinely creates 3-5 page prompts that are generating 60-100 page, very thorough reports."1
The method stacks three artifacts into one prompt:
flowchart LR
A["ChatGPT writes a<br/>Deep-Research prompt<br/>best-practices guide"] --> D
B["ChatGPT role-plays an<br/>investment analyst,<br/>lists the metrics/steps"] --> D
C["Upload the raw docs:<br/>5x 10-Ks + the S-1"] --> D
D["Combine into one<br/>Deep Research prompt"] --> E["60-100 page<br/>investment thesis"]
The analyst persona is deliberately naive — "I want to see how AI will perform with no additional investing domain knowledge" — to test the democratization claim: can someone with no finance background get an institutional-grade analysis?1 The template itself is generic ("you can see that this prompt will now work for any company"), which is why the user flagged the entire prompt rather than the DoorDash conclusions.1
Prompting is the real skill
A recurring theme across the tweets: raw ChatGPT is underwhelming; prompted ChatGPT is transformative. McKay Wrigley's demo exists to show that "Prompt tricks pay off" — the same model produces "night and day" software output depending on the prompt.5 Rowan Cheung's pitch is blunter: "ChatGPT is a free money printer. But 99% of people don't use it properly" — the hook for a bundle of 500+ prompts and 100+ business ideas.9
ChatGPT is also used to prompt other AIs. Nghia Tran asks it to write a better Midjourney prompt for a bird app icon and is "mind blown" by the result4 — the LLM as a meta-tool that improves your inputs to image models. Charlie Ward's no-code build chains ChatGPT with Replit and Midjourney to ship a working radio player with "no coding experience."6 The pattern: ChatGPT sits in the middle of a toolchain, translating human intent into whatever the next tool needs. For the broader technique, see AI Tools for Knowledge Work and Prompting.
Where it fell short
The collection is honest about limits. The MileLion's award-travel test is the clearest miss: the human expert lists genuine sweet spots (Turkish Miles&Smiles to Europe at 45,000 miles, Alaska/JAL to Japan at 25,000), but the flagged complaint is that ChatGPT "missed the Singapore to Middle East/Africa sweet spot, since 56,500 miles for a one-way Business Class ticket from Singapore to Cape Town... is probably one of KrisFlyer's best redemptions."310 The subtitle of the piece is the verdict: an AI chatbot answering all your miles questions is possible, "but not today." Domain expertise still beats the generalist on the specifics.
The Nintee retrospective is a different kind of cautionary note — a product built on the wave. GPT-3.5's launch is precisely what convinced the team it could "scale what we were already doing with weight loss" coaching11, but two years of AI-coach pivots ended in shutdown because "Digital interventions could only do so much in positive real life habit building."11 ChatGPT lowered the cost of building; it did not, by itself, solve retention.
The skeptic's counterpoint
Deliberately saved alongside the playbooks is the case against the hype. The Emily Bender profile is the anchor: her mission is to deflate "AI," a word she uses only in air quotes and says "should really just be called automation."12 Her coined term for the chatbots from OpenAI and its rivals is "stochastic parrots" — a system "for haphazardly stitching together sequences of linguistic forms... without any reference to meaning."12 She warns the tools are "born flawed" from biased training data, that there is "no view from nowhere," and that AI companionship is "nothing but placebo."12
Gabriel Valdivia's "Moves" essay carries the emotional version of the same worry — AI as a destabilizing force on white-collar work — and answers it with farmer Tim, who "sits still, surrounded by movement," unaware of Sora, "OpenAI's latest deepfake machine," and untroubled by whether he has vibe-coded a solution for his dairy farm.13 The collection holds both the tool and the critique of the tool at once. This tension is developed in AI and Irreplaceable Human Work.
The idea-generation loop
The most recent flagged use (June 2025) closes the loop back to OpenAI itself. Riffing on Sam Altman's post about the "era of the idea guy," Greg Isenberg's playbook opens by pointing ChatGPT at itself: "ask chatgpt: 'give me 10 tedious workflows a [job title] does that ai could automate'" and "scroll your own chatgpt history. if you're repeating the same prompt weekly, that's a product hiding in plain sight."8 ChatGPT becomes both the thing you build with and the mine you dig ideas out of — a fitting summary of its role as the collection's default LLM. See Startups, Indie Hacking, and Business Strategy.