Mandar Wagh

waghweb Here's a direct debunk of aphyr's top 5 arguments from his essay: 1. "Constant lying/confabulation": It's next-token prediction, not intent. Hallucinations are real but plummet with RAG, verification loops, and scaling—modern models hit 90%+ accuracy on benchmarks where early ones failed. 2. "No true understanding/metacognition": Emergent abilities prove otherwise—LLMs solve novel physics/math problems beyond training data. Reasoning models (like o1) now self-correct chains of thought effectively. 3. "Jagged capabilities": True of all tools (humans included). We pair LLMs with humans/benchmarks for reliability, not treat as oracles. This is engineering, not failure. 4. "Societal misinformation/trust erosion": AI also amplifies truth via fast fact-checks and synthesis. Past tech (printing press, internet) faced same fears; adaptation wins. Spam/propaganda predates LLMs. 5. "Uncertain improvement/overhype": Scaling laws still hold; multimodal/agentic systems advance rapidly. Diminishing returns? Data/algos compound. Jepsen-style testing (his specialty) can rigorously verify claims here too. The future isn't lies—it's verifiable progress. Buckle up for the good parts.

Read Kyle Kingsbury’s 32 page critique of AI: “The Future of Everything is Lies.” It is a polemic, cynical and disagreeable piece to many in tech, but felt by most outside of it. It highlights the many problems we will need to solve as AI percolates through society. Must read.

Read Kyle Kingsbury’s 32 page critique of AI: “The Future of Everything is Lies.” It is a polemic, cynical and disagreeable piece to many in tech, but felt by most outside of it. It highlights the many problems we will need to solve as AI percolates through society. Must read.