Worries about training data haunt generative AI like a vengeful ghost. Google has satisfied European regulators about its Bard data, to the extent that it has now released Bard in the EU. But ChatGPT owner OpenAI faces multiple data-related lawsuits and a Federal Trade Commission investigation.
Publishers are increasingly unwilling to let everyone use their content for free. While most recent discussion has been on payment for news links by Google and Meta, there’s also growing interest in ensuring that AI systems compensate owners whose content is used as training inputs. OpenAI has gotten a bit ahead of the game in a deal with Associated Press, gaining a license to use its stories in return for undisclosed fees and technology access.
Problems are also opportunities. Anthropic, which has raised a cool $1.5 billion, just launched its latest large language model, Claude2, promising it will explain its results and avoid harmful outputs. Meanwhile, website search platform Lucidworks announced it will validate results from major generative AI platforms before delivering them to users. Expect more vendors to let companies use their own data to train large language models.
RudderStack, which styles itself “the AI Data Cloud Native CDP”, has unveiled a suite of Data Apps running on its Customer 360 solution, which in turn runs within Snowflake AI Data Cloud. The apps support the fast delivery of data projects — for example, attribution data and propensity scores — from data teams to business stakeholders. They are now available to Enterprise Tier RudderStack customers on Snowflake.