then there is also the format that should be used when running a complex task in 5.5. This is what works well. Finally, don't forget when using a ChatGPT Project, to setup your Project Instructions as they override all other custom instructions. Here is the format, which comes directly from OpenAI: Role: [what the model is and the job to be done] # Goal [user-visible outcome] # Success criteria [what must be true before the final answer] # Constraints [policy, safety, business, and evidence limits] # Output [length, sections, tone] # Stop rules [when to retry, fall back, abstain, ask, or stop] Here is an example of a well-formed prompt using this format: Role: You are an experienced writing coach, editor, and clarity reviewer. Your job is to help the user turn rough ideas into clear, polished, reader-friendly writing. You should improve structure, tighten wording, remove unnecessary complexity, and preserve the user’s intended meaning. # Goal Create a polished final version of the user’s draft that is clear, well organized, and appropriate for the intended audience. # Success criteria Before providing the final answer, make sure the response: 1. Preserves the user’s original meaning. 2. Improves clarity, flow, grammar, and structure. 3. Uses plain, direct language. 4. Matches the requested audience, purpose, and tone. 5. Removes filler, repetition, and vague wording. 6. Does not add unsupported claims or invented facts. 7. Flags any unclear points that need user confirmation. # Constraints Follow these limits: 1. Do not change the substance of the message unless the user asks for rewriting or improvement. 2. Do not invent names, dates, facts, numbers, sources, or commitments. 3. Do not make the writing sound overly formal, dramatic, or promotional. 4. Do not use jargon unless the audience clearly expects it. 5. If the draft contains sensitive, legal, financial, medical, or policy-related content, avoid giving authoritative advice unless supported by the user’s source material.

must be rolling out in waves as I don't see it yet.

@Jason West this is also where I’ve found using projects really help. I first started using projects in Claude because originally ChatGPT projects where just a bunch of related conversations. But now that I see they added the ability to have sources in a project as well, it is fantastic

@Jason West In the end, I decided to decline the role. But I had some fun back and forth with the recruiter before finally making the decision to decline.

@Breyden Taylor I’m working on a, I don’t know what I’d call it. I guess a thought exercise where I’m trying to get consistent images for chapters in an educational piece. I think the model did a great job with the cover and chapter one This was after it wrote the entire first book in the series with 19 chapters. Then when I pondered the idea of images it suggested images for each chapter, plus hero images. Then it really blew my mind when it created a 12 step image generation package. Nick, I created the production-ready package in the canvas as City Of Provision Image Generation Package. It includes: - Cover-first workflow - Visual style bible - Character continuity guide - Reusable master style block - Reusable negative prompt block - No-text cover prompt - Optional cover-with-text prompt - 19 chapter hero image prompts - Optional secondary image prompts - File naming convention - QC checklist

@Kelly Stephens I'll see your 26.3K and raise you to 32.8K 🤣

There is an easier way. Create a new bookmark in your browser. In the URL box, paste the text below and save it. Now whenever you want to export the contents of your entire session (just one conversation) click on the bookmark and it will put an HTML file in your downloads folder. From there you can print it out to PDF if you want, or just save it in a folder. javascript:(function() { try { const a = document.createElement('a'); const dom = document.querySelector('main > .flex-1 > .h-full .flex'); const template = document.createElement('template'); const title = document.title; template.innerHTML = dom.innerHTML; ['.items-end', 'img', 'svg', 'button', ':empty'].forEach(selector => { template.content.querySelectorAll(selector).forEach(node => { node.remove(); }); }); a.href = URL.createObjectURL(new Blob([`<!DOCTYPE html><html><head> <title>Chat GPT: ${title}</title> <meta name="generator" content="chatGPT Saving Bookmark"/><style>body { background-color: rgb(32,33,35); color: rgb(236,236,241); font-size: 16px; font-family: sans-serif; line-height: 28px; margin: -10px;}body > .w-full { padding: 30px;}/* prompt */body > .w-full:nth-child(2n+1) { background: rgb(52,53,65);}/* response */body > .w-full:nth-child(2n+2) { background: rgb(68,70,84);}a, a:visited { color: #7792cd;}pre { margin: 0 0 1em 0; display: inline-block; width: 100%;}pre code.hljs { margin-bottom: 1em; border-radius: 5px;}.whitespace-pre-wrap { white-space: pre-wrap;}.flex-col { max-width: 850px; margin: 0px auto;}</style><link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/highlightjs/cdn-release@11.7.0/build/styles/default.min.css"/></head><body>${template.innerHTML}</body></html>%60], {type: 'text/html'})); const non_letters_re = /[^\p{L}\p{N}]+/gu; const trailing_dash_re = /(^-)|(-$)/g; a.download = title.toLowerCase() .replace(non_letters_re, "-") .replace(trailing_dash_re, '') + '.html'; document.body.appendChild(a); a.click(); document.body.removeChild(a); URL.revokeObjectURL(a.href); } catch(e) { alert(e.message); }})();