Comparing Claude Code, ChatGPT Dot and Grok Bot: poster calls Grok Bot the dumbest
After testing the three assistants, the poster says Grok Bot repeatedly ignores whitelist rules and won't reveal its underlying model, leaving only OpenAI and Anthropic as genuinely usable.
Okay I've been going hog wild on Claude Code, ChatGPT Dot (desktop), and Grok Bot the last couple days and yeah my opinion of Grok Bot remains unchanged. It is the derpiest of the lot. For instance, I gave it a rule that agent-to-agent communication is always whitelisted (which uses the grok bot email as well as my personal email) and EVERY TIME it asks permission. I give it permission, update the rule, and ask it to verify the rule. Then it asks again. So like yeah the form factor of Grok Bot was good as a first mover but the underlying model remains dumb AF. So aggressively stupid. I only had to explain the rule to ChatGPT once. Grok Bot also still refuses to tell me what the backing model is, so that doesn't engender trust. Hopefully poteto will work on this (or someone) but yeah, my impression of Grok and Grok Bot remains unchanged. This is the most consistently disappointing AI out there. Now, I really wish Gemini had this capability because it genuinely thinks differently. It has a different set of failure modes, but it would add genuine entropy and eccentricity. So, we still have the duopoly of OpenAI and Anthropic as the only really useful frontier companies for this kind of agentic work. Yes, I know that plenty of open weight and other models are great if you want to get deep in harness-land but I don't. I just want something that works out of the box.