🚀 New Video: Be CAREFUL when using ChatGpt now!
Not a "ChatGPT is bad" video. The actual complaint is narrower: a model can give a polished answer, follow individual instructions, and still lose sight of what the work was supposed to accomplish, so you end up supervising the direction of the work, not just checking the result. That's been Nick's experience with the newer OpenAI models on longer Codex tasks specifically, stated as his experience, not a universal claim.
• The one question that actually matters: not "which model is smarter," but did it preserve what you asked for, did it finish the work, and how much attention did you spend getting it there. An impressive first answer doesn't settle that if the next few turns keep missing the point.
• Public reports are treated as anecdotes, not proof: a Reddit thread about GPT-6 running extensive checks and coming back with the task unfinished is "their report, not a run I've independently reproduced," and positive reports (a successful overnight task with lower usage) get the same treatment, not cherry-picked out.
• The fair test, so you don't have to take anyone's word for it: take an actual unfinished task, give an alternative model the same information and the same goal, and check whether the important requirements survived and whether the result is usable, separate from supervision cost.
My take: this is one of the more carefully-hedged opinion videos on the channel, and worth noting exactly where it slipped. The opening line adds "and you probably want to be as well," which is a stronger universal claim than anything else in the video argues for. And near the end, Nick says directly on camera that the video carries some bias and fallacies of its own while trying to be as reasonable as possible, that's not scripted hedging, that's him saying it plainly.
📎 Full guide PDF pinned below: the three judging questions, how every public report is framed (and its limits), the usage-reset nuance, Opus 5.5 vs Astra's actual split benchmark result, and the five-check fair test to run on your own work.
👉 Which actual job has a model helped you finish lately, and where did you still have to step in?
Watch the video here ▶️ https://youtu.be/4A7uzDwAlN8
7
1 comment
Nick Puruczky
8
🚀 New Video: Be CAREFUL when using ChatGpt now!
AI Accelerator
skool.com/systems-to-scale-9517
A community for mastering AI, Agents, and Automation.
Join today for immediate value!
Leaderboard (30-day)
Powered by