I'm often asked which AI Chatbots (if any) are worth paying for. Currently, at Camino we have subscriptions for three different services…
1. Copilot - integrated into Microsoft 365, so helps directly on Word documents, and takes notes in Teams calls. The Copilot chatbot is based on GPT-4 so very capable, but it feels tuned to be less risky or creative, regularly refuses requests, and reminds you it's an AI all. the. time.
2. GPT-4o - despite being announced and partially released in May, the true "omni" capabilities - where it can use writing, speech and vision all at the same time - are not available for general use. However, even for just text input, GPT-4o is much faster and more accurate than the previous non-o version. The ChatGPT Plus subscription gets me voice access, which is great for asking random questions about disease areas or helping with analogies for school revision, and access to the DALL-E image generator.
3. Claude 3.5 Sonnet. Sonnet is the "mid-size" model from Anthropic, and the current best-in-class LLM. Claude Haiku is a smaller, faster model and Claude Opus is larger. Opus should be better, but it's currently only at version 3, and just isn't. It's painful how complicated they're making these naming schemes!
However, 3.5 Sonnet is my day-to-day AI chatbot for rewriting copy, coding, data extraction, nearly everything. Its tone-of-voice sounds more natural than GPT and it gets the gist of what I'm after much quicker. Right now, the free version has low usage limits, so it's worth the Pro subscription, which also gets you the recently added "Projects" feature.
Projects are similar to what ChatGPT calls "GPTs" where you organize a conversation around a set of files and previous content. My main one is a "Camino" project with company background, creds, website copy and our core values so any new conversation I start in this project already has the background knowledge. This morning, I asked "Do we have a policy around 2-factor authentication requirements?", which brought up where I'd briefly mentioned 2FA in another document. Claude helped me rewrite this into a proper policy mentioning the main services we use and how to configure them in about 10 minutes.
(For our client projects, we use the API developer accounts - which are separate from a "Pro" subscription - of both GPT and Claude. These interfaces are similar enough that we can switch between them depending on project and client requirements. We've tried to use Google's Gemini here - it has a larger context window - but we've not seen any benefits and a few downsides.)
Today, if I only paid for one AI chatbot it would be Claude Pro, as it's just the best language model for most uses. However, we'll still subscribe to Copilot, as it saves my team a lot of time in call notes and just by being "right there" in the Microsoft products. ChatGPT Plus is currently at risk of losing our custom… but that will probably change with whatever version they release next!