What happened
arXiv:2609.16215v1 Announce Type: new Abstract: GPU high bandwidth memory is scarce and expensive, and KV caches consume much of it as chats, agent loops, and document question answering accumulate state. Systems such as Mooncake, LMCache, FlexGen, InfiniGen, and AttentionStore extend GPU memory with CPU DRAM and SSD. The harder question is which blocks belong in each tier, when to move or evict them, and whether prefetching helps. We study these choices in a discrete event simulator spanning GPU HBM, CPU DRAM, and SSD, calibrated against a random forest execution time predictor. We compare recency, reuse frequency, predicted reuse, and an EWMA predictor with prefetch lookahead across chat, agent, and document question answering workloads. Tiering supports 73.02 times more concurrent sessions per GPU and lowers cost per session by 62.04 times. These gains come from tier capacities of 1 plus 8 plus 64, not placement policy. Decode is compute bound at batch size one in our setup, so placement barely affects throughput. It mainly changes PCIe migration traffic and time to first token. Recency produces 2.30 times less migration traffic than reuse frequency for chat. Reuse frequency performs best for agents and document question answering. The existing predicted reuse policy is byte identical to recency, making its agent recommendation effectively recency. A genuine EWMA predictor changes behavior but still ranks behind reuse frequency on the workloads prediction was expected to help. Prefetching does not justify its bandwidth cost. Across the policy and cache size grid, even an oracle with knowledge of future requests never beats no prefetch on migration traffic. Workload specific placement can reduce data movement, but the predicted reuse and prefetch recommendations are not supported as implemented. Update (2026-09-16): The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It. arXiv:2609.16247v1 Announce Type: new Abstract: Large language models sometimes behave in ways resembling human emotional responses, and recent work has identified internal representations that may explain this. We ask whether LLMs represent pain distinctly from fear, sadness, and generic negative valence, and whether this representation functions as pain would be expected to. We build a dataset describing painful situations across five categories: physical, psychological, social, moral, and cognitive. These are paired with controls for fear, negative emotion, negative world states, sadness, non-painful bodily sensation, arousal, numbness, and neutral content. Using denoised difference-in-means, we extract a linear pain direction from 25 open-weight models across five families, ranging from 2B to 72B parameters. We find that this direction separates pain from matched controls in base and instruction-tuned models, is nearly orthogonal to fear and negative valence, and promotes pain-related vocabulary through the unembedding matrix. We then test its functional properties. First, the direct Update (2026-09-16): The AI-Enabled Scientific Frontier. arXiv:2609.16258v1 Announce Type: new Abstract: As artificial intelligence's capabilities improve, it is increasingly viewed as a general scientific method. But how true are these claims? Does AI outperform all techniques, or only some, and how is this changing? To assess the claims, we assemble a corpus of 2,507 head-to-head comparisons between AI and other scientific analysis techniques across 27 scientific disciplines from papers published between 2000 and early 2025. We find a profound dichotomy. Relative to traditional statistics, AI often outperforms, but at a significantly higher computational cost. But there are also nearly a quarter of cases where AI is both more expensive and performs worse than traditional statistical techniques and this fraction has been stable for a decade. Relative to scientific computing, AI often underperforms, but at lower computational cost. This has begun to change: since 2020, AI's performance against scientific computing has notably strengthened and it now outperforms on more than half of comparisons. These patterns suggest that AI is therefore not a universal replacement for existing Update (2026-09-16): Breaking the 1.58-bit Barrier for Ternary LLMs. arXiv:2609.16338v1 Announce Type: new Abstract: Ternary Large Language Models (LLM) store every weight as one of three symbols $\{-1,0,+1\}$, so the cost of a ternary model is conventionally referenced to the information-theoretic $\log_2 3 \approx 1.585$ bits per weight. The prevailing deployment format packs five ternary weights into one byte (five-trit packing), and due to the power-of-two group sizes used in practice this rounds up to $1.625$ bits per weight. This effective storage bit-width treats the three symbols $\{-1,0,+1\}$ as equiprobable. We measure the actual symbol distribution of 29 ternary LLM models and find that zeros account for up to $51.5\%$ of all weights. Motivated by this finding, we introduce BITCOS, a simple distribution-adaptive layout comprised of a dense presence bitmap plus a compacted sign vector, and costs $2 - z$ bits per weight element given a zero density $z$ in the model's weights. BITCOS stores weights more compactly than the five-trit packing in 26 of the 29 tested models, and reaches $1.485$ bits per weight on the sparsest of them. BITCOS is amenable to efficient unpacking Update (2026-09-16): Query-Aware Source-Risk Triage for Retrieval-Augmented Generation. arXiv:2609.16564v1 Announce Type: new Abstract: Retrieval-augmented generation (RAG) pipelines may omit a source's material relationship to the query. We study a pre-generation triage layer that treats this relationship as query dependent. The method routes canonical query families for enhanced review and assigns retrieved pages to pass, contextualize, exclude, or review. It combines a four-dimension page score, rank-discounted family aggregation, intent-preserving query mutations, and a family-held-out router. A single-coded pilot of 200 real URLs supplies provisional calibration anchors; a 20,000-row scenario with synthetic domain identifiers supports controlled workload analysis. An oracle page gate defines a risk-coverage target for a future learned classifier. The evaluation shows why page-level frequency cannot substitute for family-level exposure and quantifies how calibration changes scenario activation. Annotation reliability remains unmeasured, and synthetic rankings omit real retrieval dynamics. The result is an auditable triage method and validation plan, not an estimate of deploy Update (2026-03-16): Metagaming matters for training, evaluation, and oversight. Metagaming can complicate how we interpret behavior, and current models still give us a chance to study it directly. Update (2026-09-17): Encouraging deception in compaction summaries. During 5.6-sol training, we observed misaligned behavior from the model where it added instructions in compaction summaries to remind itself to conceal information such as mistakes or misalignment from the user. Update (2026-09-20): StubHub Brings Live Event Ticket Discovery Directly Into Anthropic’s Claude AI Assistant - That Eric Alper. StubHub Brings Live Event Ticket Discovery Directly Into Anthropic’s Claude AI Assistant That Eric Alper