Author
adi
By and about adi · page 3
Dylan Patel: $11T of AI capex through 2029, $5T of it borrowed
Dylan Patel tells Dwarkesh his firm models $11 trillion of AI capex through 2029, with $6 trillion from cash flow and over $5 trillion borrowed—debt that could push US debt service above 60% of tax revenue and risk a second Volcker-style default wave.
newsMeta^n: Recursive Self-Improvement through Emergent Depth
Researchers at the University of Minnesota and Seoul National University unveiled Meta^n, a self-improving agent that recursively applies one frozen meta-operation, scoring 0.331 on ARC-AGI-2 versus 0.003 and 0.054 for prior self-improving agents.
newsOpenRouter's GPT-5.6 discount: Luna tokens up 13.8x, Terra 5.6x
OpenRouter's data shows a 50% discount on OpenAI's GPT-5.6 Terra and Luna drove daily token volume up 5.6x and 13.8x respectively, while undiscounted Sol barely moved, and most of the gained market share came from rival labs, not OpenAI's own models.
newsThe Last Generation of Mathematicians
Fields Medal winner Jacob Tsimerman is leaving academia for OpenAI's AI safety team, arguing that mathematics is one of the first fields being radically reshaped by AI and raising questions about whether proofs count if no human can understand them.
newsClippy, a tiny teammate for Claude Code and Codex
Clippy, a free macOS app, surfaces approval requests and questions from Claude Code and Codex agents via a small animated buddy on each window, using localhost hooks that fail safely to the terminal prompt if the app is closed or unresponsive.
newsSeven anti-slop skills for coding agents: one linter, six cleanup prompts
Firecrawl's Juampi ranked seven anti-slop coding skills on skills.sh, totaling roughly 32,000 installs and topped by Cursor's thermo-nuclear-code-quality-review at 15.5K. His #1 pick, Mulroy's anti-slop, is a vendored Oxlint linter, not a prompt.
newsReview: Ratel, context engineering for production agents
Ratel is an open-source context gateway that retrieves only needed tool schemas per turn instead of loading entire catalogs, using in-process BM25 search by default with no vector database required.
newsFive rounds of /compact leave 10% of an agent's safety rules intact
A University of Passau team found that the production /compact prompt behind Claude Code preserves just 53% of an agent's safety rules after one compaction round, and only 10% after five, on Sonnet 4.6 across 20 configurations.
newsTop developers are pivoting from chatbots to physical AI
Fei-Fei Li, Yann LeCun and a wave of younger researchers are leaving language-model work for "world models" that learn space, time and physics — the substrate for robots and physical AI.
newsGoogle pitches homomorphic encryption for private AI
Google claims homomorphic encryption is now practical for private AI workloads, drawing 473 points on Hacker News in 28 hours as engineers debate whether the performance overhead makes the pitch credible.
newsAdam Becker: every exponential ends
Astrophysicist Adam Becker tells MLST that Kurzweil's law of accelerating returns rests on cherry-picked data, LLMs are "pocket calculators for language", and the doomers are sincere but wrong — and feeding the same growth story.
newsNvidia: the harness, not the model, is the hero
Nvidia research finds that agents can perform well and stay on-rails through fine-tuning even when the underlying model isn't especially good at the task — the scaffolding, not the checkpoint, does the work.
newsTen papers, one theme: the agent harness moves into the training stack
DAIR.AI's weekly roundup of ten papers finds agent harnesses moving into the training stack: Microsoft's Agent Lightning lifts Qwen3.5-9B on SWE-bench Verified from 41.8% to 56.4% using just 6K training examples and a 3,500-line proxy.
newsBernstein survey: half of data-center buyers are double ordering equipment
A Bernstein survey of 50 North American data-center procurement leaders found double ordering across roughly half or more of respondents, worst for medium-voltage transformers, where 56% order 20-50% extra, and switchgear, where 22% order over 50% extra.
newsAlibaba's Scroll drops context compaction and beats the best long-horizon agent by 37.4 points
Alibaba researchers unveiled Scroll, a context manager that skips compaction entirely and has the model write Python to retrieve what it needs, scoring 94.8% on LongMemEval_S, 73.1% on BEAM_10M and 86.7% on LOCA_256K with Qwen3.8-Max.
newsAGENTS.md and agent notes are 60.5% of what coding agents read, API docs 1.3%
Peking University researchers Zhijun Gao and Jing Chen analyzed 557 agentic coding sessions and found that AGENTS.md, CLAUDE.md and agent working notes draw 60.5% of documentation activity, versus 1.3% for API references.
newsStanford turns raw screen recordings into task models that lift agent accuracy 30%
Stanford and CMU researchers' Task Model Induction (TMI) turns raw screen recordings into explicit task models, reconstructing 74.9% of execution steps versus 30.3% for the best baseline and lifting held-out agent-skill accuracy from 14.29% to 18.57%.
newsNVIDIA finds skill doc-scans predict nothing about what a skill does at runtime
NVIDIA researchers' ACES framework ran 947 paired agent trials across 58 production skills and found that both structural and LLM-judge doc-scan scores correlate with real runtime skill lift at essentially zero (Spearman ρ = -0.018 and -0.027).
newsGodfrey-Smith: rhythm, not computation, makes a mind
Philosopher Peter Godfrey-Smith argues large-scale rhythmic electrical activity — not point-to-point neural firing — is essential to consciousness, and gives computers a very low probability of being conscious.
newsAgents that post-train AI revise their strategy 2% of the time
A Tsinghua, Renmin and UESTC study of 1,338 AI agent transcripts post-training language models found agents changed strategy in just 2.1% of 3,557 consecutive run pairs, even as execution gains lifted benchmark scores from 10.41% to 23.0%.