Jesse Silver

Jesse Silver co-founded a creator-chatbot business with Adam Scrivener, building software that conducts one-to-one fan conversations and sells digital content on creators’ behalf. In his May 2024 Latent Space interview, he explained the engineering challenge as maintaining a creator’s persona across long conversations while pursuing commercial objectives. By March 2025, he was identified as co-founder of Whisper.[1][3]

The business began with customer interviews in January 2023, when Silver was looking for an LLM-based company to start. A creator friend’s time-consuming fan conversations suggested an opportunity for automation; he interviewed her and her friends, then developed a prototype with the friend who became his co-founder. They pitched it at AVN, a major adult-industry fan conference, and began bootstrapping with participating profiles. Silver recalled relaunching a more developed product in December.[1]

Silver distinguished sounding like a creator from delivering the experience fans expected. He recalled a deployment on a successful creator’s profile that produced very little revenue: the bot’s friendly introductions contradicted her established, inaccessible persona. “We're puncturing the fantasy that the creator is inaccessible,” he explained. The failure led the team to account for the creator’s brand and expected interaction style, rather than simply making every bot accommodating.[1]

An unnamed creator described the complementary problem in Jeremy Kirshbaum’s July 2024 interview. Silver gave her an on/off control, and she closely monitored the bot during its first month, stopping it and reporting responses that did not fit her character. He adjusted its parameters in response. Together they reduced aggressive selling so it spent longer building connections before proposing a video sale; she reported that sales improved afterward.[2]

The team prototyped with OpenAI and then moved to open-source models. Silver cited the cost of very long conversations, restrictions that repeatedly obstructed the workflow, and a recognizable OpenAI style in generated text. He described separate modules for chatting, understanding media and reasoning about which content to send, potentially using different fine-tuned models. State-machine elements distinguished ordinary conversation from sexual interactions.[1]

Low-code tools helped the team explore its early workflow, but adaptations across creator profiles became difficult to manage. Silver described moving to DSPy to keep the workflow in Python, make it modular and composable, and support ongoing fine-tuning. The team also used optimization of prompts and examples to draw on successful interactions from a creator’s profile.[1]

Memory had to support taking over existing conversations, including customers who had already spent substantial sums. Silver described embedding and summarizing information about preferences, purchases and earlier sessions, supplemented by heuristics. Remembering a fact did not by itself determine the next response: the workflow also had to decide whether to continue a topic or steer the conversation toward another interaction.[1]

Silver’s team combined labeled state-transition tests, targeted behavioral evaluations, representative “golden sets” for major changes, model-graded checks and human review of execution samples. He emphasized that judgments needed persona context: a fan expressing displeasure might be participating in an experience built around punishment. Monitoring changes in rates and workflow transitions, alongside inspecting conversations, helped distinguish that context from a developing problem.[1]

For prompt injection and transaction safety, Silver described multiple input and output checks, model-based sanitization, partly separating pricing from response generation, and hardcoded price guardrails. He acknowledged both exploitable inputs and instances of media being sold too cheaply. Asked whether he had solved prompt injection, he answered: “No. But I don't want to claim that I've basically solved anything as a matter of principle.”[1]

The creator interviewed in July described a Whisper Inbox that collected conversations requiring human attention, including sensitive exchanges, dissatisfied customers and questions about real-world events the bot could not know. She used those conversations to give Silver further feedback. The handoff depended on her follow-through; she also acknowledged sometimes leaving messages unanswered.[2]

Silver framed increasing creator income as the product’s central purpose and preserving their brand as a responsibility. His interpretation of fan interactions was commercially explicit: “As long as we can keep their disbelief suspended, then we can sort of make the fan happy”. He also credited creators’ expertise in their own audiences and their continuous corrections. In May 2024, he expressed interest in a future NSFW product for women offering a contextual experience sustained over a long period.[1]

In March 2025 reporting on Whisper’s $1 million raise, Silver described the intended next step: “This funding allows us to refine our behavioural cloning models and expand our AI capabilities to better support creators in monetising their content efficiently.”[3]