How good is the startup advice you get from ChatGPT, Claude and other LLMs?by Derek Lidow We are frequently asked: “how good is the advice you can get from an AI LLM?”— Our answer: AI is a decent assistant but it is definitely not a good partner or a good mentor. So, this week we want to focus on AI’s deficiencies in being a good partner or advisor. We are very aware of what kind of startup advice you can get from AI because we spend hours testing the quality of the advice it gives to many different personas and with many different prompts. We have set up development frameworks to train various AIs to be great entrepreneurship coaches. And we check on the answers they give weekly because the models are changing constantly. The challenge is that the big AI models we use in our tests, most recently ChatGPT 5.2 thinking, Claude Opus 4.5, and now the impressive Claude CoWork give you generic task-level advice based upon how they have been trained. This is problematic for four reasons: Generic advice delivers the entrepreneur no competitive advantage, which is death for a startup. AI models are trained to tell you with answers you will like. Being a successful entrepreneur always requires someone to transcend what they think they know and want to know. Good coaching requires asking tough questions. AI is not cautious with their advice, at least as of now. In every test, including with different prompts we’ve given, the LLM models have told the test persona to launch their business ASAP, in very specific ways that may not be consistent with what would be best for any corresponding real individual given their possible motivations (sometimes even putting all their limited life savings into it). And the specific instructions would yield no long term competitive advantage. As everyone’s entrepreneurial journey is unique, the key to delivering good advice requires probing for the specific issues and challenges an individual is facing at that instant. We have found that no amount of prompting and training can get ChatGPT or Claude to probe to discover and understand all the potential dimensions that define the entrepreneurial context of a specific situation.
Neither are LLMs yet good at probing for the specific characteristics of an entrepreneur, including their strengths as well as weaknesses that would require specific mitigations, beyond what the entrepreneur inputs about him or herself. We do find AIs to be mostly very good at giving transactional advice on how to accomplish specific tasks such as: editing copy, where and how to find contractors with specific skills, as well as summarizing strengths and weaknesses of competitors. We use them all the time as our assistants, but not as our coaches. We will discuss this in greater depth next week. And we will keep you updated on our successes and failures in training AIs to give good entrepreneurial advice. With gratitude, Derek and Ben P.S. If you like what we’re doing, please consider recommending this newsletter to another person you think would benefit from us. It really helps us build out our free content platform.
|
