From: aidotengineer
The adoption of open AI models is rapidly increasing, with over 50,000 models uploaded to Hugging Face per month, accelerating at a rate of more than one model per minute [00:00:08]. Notably, models like DeepSeek-R1 have surpassed closed-source counterparts like GPT-4, demonstrating that significant resources are not always necessary to compete with larger labs [00:00:22].
Primary Use Cases for Individuals
Data from FedML AI, which offers unlimited API requests to over 3,700 open AI models, provides unique insights into how these models are being used [00:00:57]. For individual users, the choice of model is often based on preference, “vibes,” or fame, rather than traditional benchmarks like MML (Massive Multitask Language Understanding) or price, as they are typically not charged by tokens [00:02:28].
Creative Writing, Role Play, and Companionship
This category, encompassing creative writing, AI role-play, companionship, and to a smaller extent, therapy and journaling, represents the vast majority (30-40%) of non-coding AI requests [00:07:32]. It also boasts the largest active user base and community, reaching tens of millions of users, including those on closed-source commercial apps like character.ai [00:12:46].
Creative Writing
AI for creative writing includes applications like NovelCrafter, a tool designed for outlining, managing, and drafting novels, facilitating high-level navigation and collaborative writing with humans and AI [00:07:44]. This segment is particularly popular with authors and the fanfiction community [00:07:56]. Additionally, some users leverage AI for creative content in games, such as Dungeons and Dragons [00:08:02].
AI Role Playing and Companionship
Apps such as Wypy Chat and S Haven are examples of platforms for AI role-playing and companionship [00:08:14]. Despite social stigma often associating this segment with explicit content, it remains the leading use case for AI today outside of code or agent workflows [00:08:20].
Contrary to popular belief, over 60% of users in this segment are women, mirroring the market dominance of romance novels targeting women [00:08:55]. Users, particularly women, tend to engage with these apps for long conversations to destress and discuss their daily lives, often due to real-life partners being non-existent or emotionally unavailable [00:09:31].
Therapy and Journaling
This category also includes therapy and journaling applications, although the usage is heavily fragmented across many dedicated commercial apps [00:10:00]. Many users also adapt existing chatbot platforms or their companion apps with a “therapy character” for this purpose [00:10:24].
The “Vibes” Factor and Model Selection
In this segment, model performance metrics like MML are largely irrelevant; the focus is entirely on “Vibes” [00:10:48]. This means models are not designed to quickly answer questions or solve problems. For instance, in therapy, the AI’s role is to guide and empower the client, not to provide direct answers [00:11:21]. Similarly, in story writing, role play, and companionship, the concept of “Slow Burn” is preferred, emphasizing the journey and entertainment rather than rushing to a conclusion [00:11:35].
This community is highly dynamic, with top models constantly changing weekly [00:11:01]. Examples of models mentioned include the “cowboy” Alind Magnum and the Rosa model for science fiction [00:12:01]. Despite the rapid shifts, users sometimes return to old favorites, likening it to playing a retro game [00:12:32]. This segment also benefits from a large number of community fine-tuned models [00:10:57].