Security is another reason to build your own custom models. Cybersecurity firm Riversafe surveyed 250 U.K. CISOs and found that around 20% have had sensitive corporate data leaked through employee use of generative AI. Without proper controls, data entered into a model like ChatGPT may be visible to other users of the tool.
Move over, Large Language Models: Microsoft thinks a Small Language Model is often a better fit. They’ve just released something called Phi-3-mini, which can easily run on a smartphone or laptop computer. This makes custom models cheaper to build and supports applications where a network connection is unavailable.
Procrastinators of the world can throw a party whenever they get around to it: Google has once more pushed back complete third-party cookie deprecation. The new target is “early next year.” Reasons for the delay include concerns expressed by U.K. data regulator Information Commissioner’s Office, an ongoing inquiry by the U.K. Competition and Markets Authority, and widespread discontent in the advertising ecosystem.
Amperity has launched Lakehouse CDP, which delivers CDP functions within “lakehouse” data stores including Snowflake and Databricks. Amperity tools for data modeling, data quality, identity resolution, generative AI, segmentation, and performance measurement will send all results to a company’s own lakehouse rather than a separate Amperity-managed database. Modules can be used independently but the real benefit here is that Amperity is one of the few “composable” CDP vendors who offer a complete set of pre-integrated CDP functions.