As we announced previously on this blog, around 1/3 of all messages sent on Wavelength invoke the AI assistant in a group. We’ve been hard at work making the assistant even more capable, and today we’re launching an open beta of a huge improvement on that front.

The AI assistant can now access the internet and current news, can perform math computations, and is aware of timezones. This means that the assistant can now answer all kinds of questions that it couldn’t before — it can tell you about recent world events, recall of factual information is much more reliable, and the bot can help you solve equations and analyze data.

Minimizing hallucinations — where the model will confidently assert something that has no basis in reality — is one of the hardest problems in AI, and we’re really excited about making significant progress on this front with the Wavelength AI assistant. Language models are probabilistic in nature and lack any source of ground truth, which means that they are prone to making stuff up when they don’t know something — and they don’t know a lot, having often been trained on data that is years out of date. We were able to utilize OpenAI’s new function calling API to give our AI assistant access to the full internet: the ability to perform search queries and to load web pages.

The result is a helpful AI assistant unlike anything currently available in a social app — the assistant uses the web to ground itself in fact when it makes sense to do so, but is also flexible, remaining friendly and conversational, and able to help with creative tasks.

With our latest app update, you’ll see an indicator when the assistant is “thinking” or “checking the web” so that you can track its progress. The assistant will include citations when it checks the web so that you can verify the answers for yourself.

We’re looking forward to your feedback! Get the update from the App Store and try sending the AI assistant a direct message or mentioning it in a group to get started.