Contiguity's 105 billion parameter, 100k context window LLM - Revolutionizing Threat Detection.
Note
LiNGO is a proprietary model developed by Contiguity, designed to remain undisclosed to the public to safeguard our interests. Public release of LiNGO would enable threat actors to preemptively test their content to avoid detection. Instead, we will provide access to LiNGO through the Contiguity Inference API starting August 2024 to select, verified users.
Spam detection has come a long way from its early days. Traditionally, companies relied on static rule-based systems to filter out harmful requests. However, since these systems used predefined criteria: (i.e) specific keywords, sender addresses, and suspicious patterns they were were easily circumvented by threat actors who quickly adapted by tweaking their requests to avoid detection.
Recognizing how far LLMs have come over the past few years, it became clear that they would be helpful in identification of text. Whether ChatGPT helped you make your email just a tad bit more professional, Gemini rewrote your whole email to make your arguments stand out, or Mixtral generated code for you, it demonstrated that LLMs could "understand" content. We're leveraging this understanding to classify texts, emails, and calls done using Contiguity's API.
Applying such a view to spam detection, a seemingly cat-and-mouse game, it is prime to be adapted to use AI.
LiNGO can analyze and interpret the subtleties of communication, making it far more adept at identifying sophisticated spam techniques while reducing false positives. LiNGO can distinguish between a legitimate business inquiry from a new client and a cleverly disguised phishing attempt posing as a request for a quote, a Chase Bank deposit confirmation and Fake Amazon Purchase scams, communications from the UK Government and Nigerian Princes, et al.
There's never been better spam detection than this.
We've been using LiNGO at Contiguity since June 2024. It has prevented over 155,000 requests, at a 0.75% false positive rate. Improving this number to <0.25% is our goal.