Anthropic's $1.5 Billion Copyright Settlement Receives Final Approval

According to TechCrunch, the roughly $1.5 billion (~1.5B) settlement between Anthropic and the copyright holders has received final approval. The case is one of the landmark lawsuits over whether copyrighted works were used in AI model training data, and it marks a milestone for the industry.

That said, the ruling resolves only this particular case. The broader, cross-cutting issue of using copyrighted works for AI training remains unresolved, and individual lawsuits and legislative responses are expected to continue. The terms of the settlement and the court's reasoning are likely to draw attention as reference material for similar lawsuits going forward.

Fable 5 May Have Disproven the "Jacobian Conjecture"

According to Mashable, Anthropic's model "Fable 5" has reportedly "disproven" the Jacobian Conjecture, a famous unsolved problem in algebra. The story also gained traction on Hacker News, with mathematicians weighing in.

At this point, it appears to still be at the stage of discussion based on model outputs, not a final conclusion that has survived rigorous verification of the disproof. Even so, it is drawing attention in the research community as a symbolic example suggesting that LLMs are stepping beyond mere pattern extraction and into the process of mathematical reasoning.

Google Vanishes from the Top 15 of LLM Leaderboards

In Reddit's LocalLLaMA community, the fact that Google-made models have completely disappeared from the top 15 of the llm-stats.com leaderboard became a topic of discussion. The post points out that Google has not recently released a model in the same class as Sol or Fable, and that its past models also fall short in terms of reliability.

As background, hypotheses include (a) the company's own shift toward on-device inference, and (b) the possibility that releases are being held up by internal organizational walls. While Apple holds the advantage on the hardware side, where Google's strategy is heading remains unclear. That said, internal models and undisclosed deployments may exist, so conclusions based on rankings alone should be avoided.

The Debate Over Whether Open-Weight AI Is Decelerationist

An essay published on Substack has sparked debate on Hacker News as well. It raises the question of whether releasing open-weight (broadly, open source) models ends up expanding safety risks by lifting AI capabilities across the board — and therefore contradicts the interests of forces aiming to "decelerate" AI.

On the other hand, some counter that open weights also promote external audits and safety research, meaning the issue cannot be reduced to a simple binary of "acceleration vs. deceleration." The open vs. closed debate appears to be growing more complex as it intertwines with safety, economic, and geopolitical dimensions.

US Politicians Lose Seats Amid Backlash Against AI

According to Fortune, voter aversion to AI is intensifying in the United States, and politicians seen as pro-AI have begun losing their seats in elections. The topic has also drawn interest on Hacker News, picking up 26 points and 13 comments.

The reporting notes that beyond anxiety about jobs, the degradation of the information landscape by AI and privacy concerns are moving voters' sentiments. Cases in which technology policy becomes a direct election issue are on the rise, and this is likely to influence future regulatory debates as well.

---