It is becoming clear that generative AI exposed to the public must be an adversarial system. From jailbreaks to prompt injection to inherent bias in the models themselves, there is no model on the market today that can be put into customer-facing production, as is, with no supervision. None. Not GPT-5, not Gemini, not WatsonX,
Whether you want them to or not, whether it’s even right or not, people are asking large language models questions about your business. And you have no visibility into this. “Hey ChatGPT, compare pricing Salesforce, HubSpot, and Zoho CRM for a company of 50 employees and annual revenue of about $50 million per year. Assume
Proving content authenticity is going to be a matter of lineage. Here’s what I mean. AI companies in consumer interfaces are adding things like watermarks & fingerprints. But when you dig into the architecture, these artifacts are added AFTER the generation. If you use the actual models themselves, these fingerprints are not part of the
How do you vibe code? First coined by OpenAI cofounder Andrej Karpathy, vibe coding (aka prompting AI to write code for you) has become all the rage. Yet people who try it get… not great results a fair amount of the time. Why? Why does it look so easy for someone like Karpathy and then
