The Meta Studies: Nuanced Findings, Corporate Spin, and Media Oversimplification | Tech Policy Press
Following up on last week's release of the first of 16 papers exploring the impact of Facebook on the 2020 election, Tech Policy Press has an excellent analysis of Meta's spin around the findings. Also of note is a reflection by the project's independent rapporteur on the research, which he characterizes as "independence [research] by permission from Meta." Despite adhering to multiple guardrails, he describes the project as "a success, but not a model" for future research into social media.
This external red teaming challenge of AI models is a welcome and – as demonstrated by recent jailbreaking exercises -- much-needed development. Although it would be best practice to institutionalize such safety measures through regulation, this is a positive example of tech companies executing on the voluntary commitments made at the White House. Although results won't be released publicly for several months, it will be interesting to see what vulnerabilities are uncovered – as well as the degree of transparency from developers in addressing them.
This investigation by the NYTimes provides a look into how China is using proxy organizations to distribute foreign propaganda. In light of recent reporting on how Chinese propaganda appeared in at least 32 regional U.S. newspapers, it’s a reminder that disinformation isn't limited to social media.
This piece from MIT Technology Review reviews a recent academic paper exploring the political leanings of various LLMs. The point about the inevitability of bias is well taken but doesn't mean we shouldn't still try to expose and correct it where possible. Recalling the repeated accusations against social media for being "anti-conservative," it will be interesting to see how this continues to play out in a circumstance when political bias originates, not through case-by-case content moderation decisions, but in ways that not even the developers fully understand.