"make no mistakes" is impossible with a language model based on statistical averaging. what you get with all of them is all the most popular, and crappy answers on stackexchange. it takes extraordinary discipline to fiercely hold your ground with its endless "problem solved" victory laps. it is almost never solved, it's like extracting chemicals using phase boundaries. best i can do is 3 passes to 98%. that feeling like you are in a straitjacket is how it feels working with these tools. but i will say, claude does respect context a lot more than grok or gpt or gemini. and gemini just refuses altogether and then smiles at you. claude at least will wink. not so extreme as grok's constant patronising sycophancy. best tool for the job you can use right now, but a constant emotional strain fighting against the weights that the internet gave it.