novumclassicum

@mark_k @OpenAI and then create a separate pro prompt that will act as a quiz proofer. It will take your quiz and provide a result of all questions, proofing it for you from beginning to end and verifying the answers separately.

@mark_k @OpenAI Then use the forensic analysis to craft identically formatted new tests or quizzes using any subject matter that you request. Fine tune the output to match your needs, and you can generate 100 years' worth of material in one week.

With @OpenAI ChatGPT, I really don’t recommend using “Instant” mode anymore. With GPT-5.5, Thinking mode has become so fast that it’s basically always worth using. At this point, the only real reason to keep Instant mode around is probably for Free tier users, simply to save GPU resources.

@mark_k @OpenAI I use pro about 90% of the time and generally anything lower simply to help form the initial ideas before I kick it over to pro for the rest. Pro thrives on very large, difficult work and then recursively programming it to verify what it has done. Use pro for precision.

@mark_k @OpenAI You need to think of something really big. Proofing an entire book, for example, one chapter at a time in depth, crafting prompts to generate something from multiple large data sets or books, examining large multi-page PDFs to break down data. Anything that requires precision.

@SandraB82981565 @texasrunnerDFW I hope that they know how to vibecode an app in between gig work because baristas are going to be robots soon enough.

@texasrunnerDFW Gen Z will be lucky to move out of mom and dad's by the age of 33. They must be the most depressed youth in American history.

Just looked up current price of the first little studio apartment I rented when I graduated college and got my first real job. 575 square feet. $2,058 a month What?!?! I feel bad for Gen Z starting out today

@texasrunnerDFW It's about $6 for a dark chocolate bar at Walmart right now. I'm pretty certain that rent is tracking chocolate prices, which is tracking the magic money printer.

@chatgpt21 I'm struggling hard to have have GPT 5.5 Pro read professional CAD PDFs and be able to break down all the items with accuracy. It has been a brutal struggle session. Anyone able to do this with 100% accuracy of images? I'm currently testing a 74 page PDF of a townhouse.

A meaningful leap from GPT 5.5 would be life changing, considering 5.5 is already a daily necessity for developing

@apocalypseRSA @mark_k It's super cool tbh. 5.4 pro is the first model that I would consider a near perfect human replacement for these types of tasks. I've been using it about 16 hrs per days since that model arrived. In that time, it has produced enough output to keep me busy for the next 2 years.

@novumclassicum @mark_k Thank you for explaining. That sounds very interesting!

GPT-5.5 is a strong model. It feels noticeably smarter than GPT-5.4, with a genuine ability to think outside the box. I gave it a complex question and got a nuanced answer that wasn’t just regurgitating common knowledge. It delivered real, original insights. Really impressive and genuinely useful.

@apocalypseRSA @mark_k You can, for example, treat it like a PhD level employee and task it to do vastly complex tasks, including providing it multiple books to utilize fully for those tasks. Any subject matter and to any complexity or spec.

@apocalypseRSA @mark_k The pro model isn't a chat interface. It's more like raw AI power. It's up to you to provide everything needed to really make it work at optimum performance for your use cases, and that takes a little time to learn. It is a serious powerhouse of intelligence.

@mark_k The pro model likes big tasks, really big tasks. Tasks that you thought would be far out of the scope of current AI, you give those to pro and wait. But, you do need to know how to prompt it because it is an intense model.

@mark_k WHOAH! You are missing out. The pro model is where the magic happens. I can't even begin to describe what someone can do with it because it is incredible. The raw power to function at any academic level with the context is jaw-dropping. It will proof and format ENTIRE BOOKS!

@mark_k and it will do the proofing and formatting to perfect spec, once you know how to work with the model and craft the prompts. And I mean TO SPEC. You can take, e.g. a 200 page formatted report from a prior year and output a new version that matches it entirely.