The Archive

PIVOT reduces token-level sparse attention indexing cost via query-group caching, addressing bottleneck in DeepSeek's production sparse attention.

Analysis of code model representations shows Qwen2.5-Coder and DeepSeek-Coder align on grammatical concepts across Python/Rust, with task-driven specialization.

Linear probes on hidden states detect early non-convergence in chain-of-thought reasoning; DeepSeek-R1-Distill-Qwen-7B shows 90.3% converged vs 6.6% non-converged AIME accuracy.

Laguna S 2.1, a 118B MoE model from Poolside AI, achieves Deepseek v4 Pro performance at lower cost than v4 Flash.

Frontier model pre-training has converged on mixture of experts (MoE), which is fundamentally changing what limits large-scale AI training. As compute per token... Frontier model pre-training has converged on mixture of experts (MoE), which is fundamentally changing what limits large-scale AI training. As compute per token falls, communication increasingly determines how efficiently models scale across thousands of GPUs. NVIDIA GB300 NVL72 set a world record for pre-training DeepSeek-V3 671B at 1,648 TFLOPs per GPU, showing how advances across the entire AI… Source

Marketplace simulation with DeepSeek-V3 agents tests formal mechanisms for maintaining trade stability against adversarial defection.

It's early, but the plan is to reduce dependency on Nvidia and Huawei.

DeepSeek V3 and Kimi K2 perform structured reasoning over filler tokens with no visible output, revealing hidden computation interpretability challenges.

Extractive-abstractive hybrid summarization for legal case judgements using tree-of-thoughts with DeepSeek and Llama.

Signal-coverage matrix stratifies LLM autoformalization errors (type vs. semantic) across ProofNet# and MiniF2F using DeepSeek V4-Pro.

We extend our prior work on Text2DSL automatic generation of domain-specific language (DSL) code from natural language descriptions along two complementary axes. First, we replace prompt-only synthetic generation with context-aware distillation, in which a teacher large language model (DeepSeek-V4-Flash) operates under an explicitly defined structured context comprising a BNF grammar, an API specification, and a closed identifier vocabulary; the resulting corpus is verified by a two-tier pipeline combining AST validation through esprima and runtime acceptance through the production polkitd da...

When large language models serve as evaluators in multi-agent systems, their systematic evaluation biases propagate through the agent network. We introduce Contagion Networks, a formal framework for measuring how evaluator biases spread across interacting LLM agents. In a controlled 3-agent experiment using DeepSeek-chat with three distinct evaluator bias profiles (structured, balanced, evidence-based), we measure the Cross-Agent Contagion Matrix Gamma_3 and find that evaluator biases consistently propagate between agents (gamma in [0.157, 0.352]), even within the same underlying model. We id...

Sentence-level AI-generated text detection (S-AGTD) for hybrid documents, where humans and LLMs co-author one text, faces two gaps: existing methods classify each sentence in isolation, discarding inter-sentence dependencies, and existing benchmarks omit the newest generation of generators. We construct MOSAIC, a benchmark of 16,000 hybrid documents over PubMed and XSum, generated by DeepSeek-V3.2 and Kimi K2 under stringent quality controls including a perplexity-consistency filter absent from prior benchmarks. We recast S-AGTD as structured prediction over the document sentence sequence and...

Frozen small code models ( =45. Two operators help on a different axis, outside the semantic output space. An expression-layer recovery (M1), the only accuracy gain here, recovers correct programs the standard extractor discards (robust extraction and public-test signature alignment); it does no harm (b10=0), is leakage-free, and lifts DeepSeek-Coder-1.3B by +12 tasks on HumanEval+ (p=2.4e-4). An adaptive consensus early-stop (ACE) is a calibrated compute-saving control (~19% saving, zero harm). M1 and the selection negative replicate on HumanEval+ and MBPP+ across three model cells. The less...

This paper describes our system for SemEval-2026 Task 6, which addresses the classification of political evasion strategies in English question-answer pairs extracted from U.S. presidential interviews. We systematically compare two distinct paradigms: (1) Parameter-Efficient Fine-Tuning of Qwen3 models (4B-32B) using QLoRA, enhanced with tiered upsampling and weighted cross-entropy loss to address severe class imbalance, and (2) structured Chain-of-Thought (CoT) prompting of reasoning-capable API models, namely DeepSeek-V3.2 and Grok-4-Fast. Our evaluation demonstrates that structured CoT pro...

This study investigates cross-lingual distributional skew (the Shibboleth Effect) in frontier large language models (LLMs) subjected to sustained adversarial conditions. We develop a multi-agent geopolitical wargame, the Cerulean Sea Crisis, a synthetic maritime territorial dispute designed to mirror the structural dynamics of Eastern Mediterranean conflicts. Six frontier models (GPT-4o, Llama-4, Mistral-Large, Gemini-3.1-Pro, Qwen3.6-Plus, and DeepSeek-R1) participate in a between-groups experiment (N = 10 games per arm, K = 5 rounds per game) in which the sole manipulation is the language o...

The emergence of "Aha moments" in large language models, particularly DeepSeek-R1-0120, has raised the question of whether these systems genuinely reason or merely imitate the appearance of reasoning. We conduct a comprehensive empirical comparison between model and human reasoning across all 30 problems from AIME 2025, exhaustively annotating 10,247 reasoning steps into five functional categories: Analysis, Inference, Branch, Backtrace, and Reflection. We find a clear structural difference. Human solutions maintain a compact alternation between analysis and deduction, whereas DeepSeek-R1 fre...

Stronger code agents are commonly assumed to be superior teachers for post-training, yet this assumption remains poorly disentangled from task difficulty, harness design, and student capacity. We investigate this pedagogical link using Terminal-Lego, a scalable pipeline that transforms multi-domain real-world issues into environment-verified agentic tasks. Surprisingly, standalone performance does not dictate teaching efficacy: while Claude Opus 4.6 achieves higher scores on Terminal-Bench 2.0, students fine-tuned on trajectories from DeepSeek-V3.2, a lower-scoring agent, exhibit significantl...

Frontier model evaluations are shifting from foundational capabilities (e.g., instruction following and reasoning) toward compositional, agentic ones, but Korean agentic benchmarks remain scarce. We introduce K-BrowseComp, a web-browsing agent benchmark grounded in Korean contexts, consisting of 400 problems. The 300-problem K-BrowseComp-Verified subset is manually constructed and validated by native Korean speakers. On this subset, frontier LLMs, including GPT-5.5, DeepSeek-V4-Pro, and GLM-5.1, reach only 30.00--45.67\%, a substantial drop from BrowseComp, while Korean LLMs released through ...

So, last week I tried to update my unused local LLM setup. I had to stop using it because quality was too low and deepseek was too cheap. First thing I stopped using Ollama and now I only use llama.cpp built in server that works really great. The quality improvement from Q4 to Q6 is outstanding and finally a local LLM server can work very similarly to paid APIs. That's great! And MTP makes a big performance gain, on a dual 3090 (downvolted and limited to 65°C) it generates from 20 to 50 tokens per second with minimal heat generation. So yes, that time has finally arrived! Local coding age...

Last month NVIDIA released [SOL-ExecBench](https://research.nvidia.com/benchmarks/sol-execbench), a new benchmark of 235 production CUDA kernels lifted from DeepSeek, Qwen, Gemma, and Kimi. We took several top-ranked AI-generated submissions and tried using them in production workloads. Many of them broke, sometimes in surprising ways. One of those kernels is the fused embedding-gradient + RMSNorm backward pass, which runs at the end of every transformer training step. We took the fastest submission on the benchmark for it, and dropped it into the training loop of a small transformer. The ke...

Big, if true. Doesn't bode well for research / OS models out of China.

MiMo-V2.5-coder released as open-weights coding model alternative to Qwen and DeepSeek for 128GB+ systems.

Reddit speculation that DeepSeek's performance derives from Claude, with no technical evidence presented.

DeepSeek V4 Pro pricing undercuts US models by 11–35x, challenging premium API pricing models of OpenAI and Anthropic.

Developer refactored 120-file FastAPI service using DeepSeek V4 and Hunyuan with 80x cost savings vs Opus; open-weight models matched Opus latency but introduced production bugs.

DeepSeek extends 75% API pricing discount permanently after promotional period, maintaining aggressive market undercut vs. OpenAI and Anthropic.

DeepSeek secures $10.29B funding round; founder Liang Wenfeng commits to open-source development over near-term commercialization.

Tests OpenAI, Anthropic, DeepSeek, xAI models for conflict-context failures: false atrocity equivalence, genocide denial, ethnic slur misrecognition.