Carnegie Mellon University

Headlines · 5

1. 19h
   Apple, CMU's SSR turns agent reasoning into multiple-choice, cutting per-turn latency 90%

   Researchers at Apple and Carnegie Mellon University propose SSR (Selection-Based Structured Reasoning), which replaces each free-form thinking step in an agent loop with a choice from a library of six predefined reasoning options. Scoring reuses the model's own token likelihoods and prefills all options in parallel, so no new tokens are generated autoregressively, with no extra classification head and no SFT warm start. On Qwen3-VL 2B/4B models with H100 + SGLang across seven multimodal search benchmarks (MMSearch, InfoSeek, SimpleVQA, LiveVQA and others), the paper reports per-turn reasoning latency down more than 90% and total per-question latency down 28–54%, while success rates match free-form thinking (4B SSR+GRPO averaging 61.37%, 2B reaching 51.26%). The authors say SSR plugs into existing SFT and RL pipelines and is best understood as reorganizing the reasoning space from open-ended language into a compact, reusable strategy space, noting that open-ended math, theorem proving and code design still need free generation.

   机器之心 · 🔥 6
2. 1d
   17-year-old Kimi researcher Nathan Chen leaves Moonshot AI for the Bay Area

   On Oct. 15, 17-year-old Nathan Chen announced on X that he had left Moonshot AI's Kimi team, describing the past year as the best of his life and thanking the team for hiring him at 16. He is a co-first author of the Attention Residuals paper, which Chinese reports describe as one of the foundational technologies behind Kimi K3, and he also worked on KDA kernel optimization, optimizer research and Tilde Research's MoMoE project. Anthropic researcher Sholto Douglas invited him in the comments to work with teams including Prime and Base10 in San Francisco; Chen confirmed he is heading to the Bay Area but replied "no" when asked whether he was joining Anthropic, leaving his next employer, role and start date unconfirmed.

   36氪 人工智能 · 🔥 24
3. 2d
   Apple's SSR recasts agent reasoning as selection, cutting per-turn latency over 90%

   Apple's machine learning research team published a paper, "Selection-Based Structured Reasoning: Toward Efficient Multimodal Search Agents," proposing SSR, a framework that replaces freeform reasoning before each multimodal agent action with a selection over pre-specified, reusable natural-language reasoning candidates scored by likelihood given the current context, without an auxiliary task head. Because the reasoning traces are pre-specified, the method uses teacher-forced prefilling to score token likelihoods in parallel across candidates via a shared context KV cache. The authors report that on seven multimodal search benchmarks with 2B and 4B models, across multiple reinforcement learning objectives and supervised fine-tuning, SSR matches the success rate of leading search agents at the same scale while cutting per-turn reasoning latency by over 90% and total per-question model inference latency by 28–54%.

   Apple Machine Learning Research · 🔥 3
4. 6d
   AI infra spending boom meets idle GPUs at Beijing salon

   A closed-door salon on AI infrastructure hosted by Touzhongwang with Zhongguancun partners, republished by 36Kr, examined the gap between huge AI infrastructure spending and weak efficiency. It cites ByteDance's $29.6 billion syndicated loan, Alibaba's HK$80 billion placement, more than $80 billion raised earlier by Google and Amazon, and NVIDIA's $500 billion plan, while noting China's daily token calls topped 140 trillion in March, over a thousand times early-2024 levels. Participants cited a Carnegie Mellon test of 756 GPUs running 31 days with the cluster waiting nearly 20% of the time, claimed over half of OpenAI's inference energy goes to idle work, and said average GPU utilization at Chinese smart-computing centers is below 30%. Several investors called lowering token cost and raising utilization the new deciding factor; SGLang's RadixArk reportedly raised a $100 million seed at a $400 million valuation, while one dissenting view argued inference revenue will fall sharply as algorithms compress token use by orders of magnitude.

   36氪 人工智能 · 🔥 0
5. Oct 9
   What mathematicians should know about Lean: reliability, soundness bugs and AI

   In a guest post on Terence Tao's blog, Thomas Hales surveys formalization and the Lean theorem prover, arguing that autoformalization became a practical reality in 2026: milestones cited include Math Inc.'s autoformalization of the 24-dimensional sphere-packing problem, Meta's ATLAS textbook project, Anthropic's 13 million lines of Lean for Fermat's Last Theorem in 11 days, and OpenAI's Lean formalization alongside its Navier-Stokes announcement. He also examines Lean's 2026 "Summer of Soundness Bugs" — one bug yielded an illicit disproof of the Collatz conjecture — which were found by security researchers using frontier models, quickly patched, and followed by re-verification of mathlib. Hales stresses that Lean proofs must be kernel-checked and then human-audited for statement fidelity.

   Hacker News · AI（100+ 分） · 🔥 0

Experience and discussion from the community

Share my Carnegie Mellon University experienceAsk about Carnegie Mellon University