Summaries by Rahul Matthan

The Delhi High Court's ruling in OpenAI vs. ANI declared AI training on protected content isn't copyright infringement, citing a fair dealing exemption for research. While laudable for its outcome, the author criticizes the court's reasoning. Instead of outright declaring training as non-infringing (akin to human learning and fact extraction), the court deemed it an infringement excused by an exemption. This creates legal uncertainty for the AI industry, as fair dealing is a defense, not a right, with limited geographic scope and a narrow interpretation. The author advocates for a clearer, invariant rule, not a precarious, conditional reprieve.

The author argues that while open-weight AI models garner hype, they only represent part of the frontier AI equation. Accessing model weights, like Kimi K3, doesn't equate to accessing frontier intelligence without significant inference and test-time compute. Modern AI demands substantial computational resources, engineering expertise, and 'process knowledge' for reasoning and tool use. Simply downloading weights provides no utility for serving advanced AI at scale. India must develop its own inference infrastructure and expertise to truly benefit, recognizing that open weights don't guarantee open access to intelligence.

India's I4C ordered GitHub to remove Bitchat, a P2P messaging app, for features preventing surveillance, not for an unlawful act. The author critically argues this misinterprets the IT Act, reversing privacy principles. He contends banning tools for potential misuse sets a dangerous precedent, comparing it to banning knives rather than prosecuting criminals. The author highlights the order's ineffectiveness due to Bitchat's open-source nature and predicts a “Streisand effect,” boosting the app's popularity. This action is deemed muddled thinking and a major “own goal” for Indian technology regulation.

AI frequently fails by solving irrelevant problems, driven by available data instead of genuine needs. Many healthcare AI tools, like radiology segmentation, are accurate yet impractical; crucial problems like triage are neglected. AI struggles with misdefined problems or ignored context. Successful AI addresses real practitioner pain points. The text advocates an inverted development: practitioners must define desired outcomes, guiding AI specialists to build solutions that meet actual needs, rather than letting data dictate the problem.

The World Cup reveals technology, intended for precision, often complicates human judgment and trust. Automation, aiming to eliminate bias, paradoxically degrades essential decision-making. Current "human-in-the-loop" systems are ineffective as people defer to machines. The author proposes inserting automation *into* the human loop: humans make initial judgments, then technology tests them. This "sequential unmasking" preserves human agency and trust. Technology should *aid* our decisions, ensuring they remain solely ours to make, fostering confidence.

Javier Milei advocates AI-run corporations, extending limited liability to foster innovation. The author critically argues this overlooks the crucial role of human accountability. Historically, limited liability worked because individuals operating companies faced personal repercussions, ensuring ethical conduct. Milei's proposal removes this essential safeguard. The East India Company demonstrates the perils of unchecked corporate power without human oversight, leading to immense suffering. AI, lacking human motivations like self-preservation, cannot be deterred by conventional penalties. Ultimate responsibility for AI systems must always reside with human creators, preventing escape from consequences.