Host Sowbhagya Suresh shares some of the latest news and cutting-edge research from across Microsoft—from understanding the challenges facing coding agents to uncovering how people use AI at work and improving extended-range weather forecasting. The post Microsoft Research Focus – Edition #97 appeared first on Microsoft Research .
LFM3 on Linux and Windows Liquid AI's [LFM3](https://ollama.com/library/lfm3) models, including text-to-speech in [lfm3-audio](https://ollama.com/library/lfm3-audio) and [lfm3-omni](https://ollama.com/library/lfm3-omni), now run on Linux and Windows as well as Apple silicon. Full Changelog: https://github.com/ollama/ollama/compare/v0.40.12...v0.40.13
Meta has launched a "tamagotchi-style" pet game called Waddle World, with virtual creatures that chat with users and create doodles to illustrate conversations. According to…
Weight-only, lossy quantization substantially reduces the storage and memory requirements for serving large language models (LLMs), but the growing key-value (KV) cache remains a bottleneck for long-context inference on personal devices. We present Group-Adaptive Additive Codec (GAAC), a lossless KV compression and serving system that exploits exponent redundancy to reduce storage and accelerate attention without introducing KV approximation. GAAC encodes BF16 exponents using calibrated per-channel base-plus-offset codes over contiguous ranges. GAAC adapts code widths across small groups while preserving KV values bit-for-bit. A fused attention kernel reconstructs KV tiles on the fly, avoiding full-cache decompression. Block-wise compression and fixed-capacity slots support incremental cache updates and paged serving. Across five tested models, GAAC achieves 1.405-1.407lossless BF16 KV compression. We also study compression performance on low-precision tensors FP8 E4M3 caches ( 1.09-1.14) and E5M2 caches (1.25-1.30). On NVIDIA DGX Spark, GAAC accelerates decode attention at a 64K-token context by 1.19-1.27and end-to-end W4A16 decode steps by 1.04-1.15. In serving, it reduces resident KV memory by approximately 26% and increases aggregate throughput by 13-29% under comparable KV-memory budgets. On Apple M4, GAAC achieves 1.02-1.25attention speedups, demonstrating portability beyond NVIDIA GPUs.
The rapid progress of software engineering agents drives interest in applying them to open-ended problems such as GPU kernel optimization. However, we find that existing coding agents converge to local optima on real-world, expert-written kernels (e.g., FlashInfer). We introduce Evolving Agent Teams, a multi-agent framework that escapes such local optima. It features (1) a central supervisor that diagnoses each local optimum and restructures a team of agents; and (2) a persistent file system that comprises a learning document, an evolution history, and per-team artifacts, grounding restructuring with accumulated experience. Under a matched token budget, Evolving Agent Teams consistently outperforms strong coding-agent baselines. With an extended token budget, Evolving Agent Teams generates a CUDA kernel that reaches 1.2x mean speedup over the production FlashInfer MLA paged decode kernel from scratch and wins on 36 of 47 production workloads from FlashInfer-Bench (0.86-0.99x on the remaining 11). It also demonstrates superior performance on tasks such as KDA prefill on AMD MI350X (3.01x over vLLM) and NVIDIA B200 (3.44x over FlashKDA), MoE expert-offload serving on Apple silicon (30.3 tokens/s decode for Qwen3.8-Flash-Next in MLX 4-bit on a 48 GB M4 Pro), and RoboDojo robot manipulation (62.1 on held-out seeds vs. 55.3 for the best leaderboard entry). Code is available at https://github.com/caoshiyi/EAT.
Today, I’m talking to Kevin Roose, journalist and author of the new book The AGI Chronicles, which details the recent and explosive transformation of the…
From Apple's repeatedly delayed Siri AI to OpenAI's messy "super app" launch, here's a look at the AI projects that shut down or missed expectations.
Ollama now supports speech models, starting with Phonon from Gradium. Read text aloud in five languages and dozens of voices, streamed as it is generated, over 100 times faster than real time on Apple Silicon.
Lab Director Chris White has worked on research challenges with real-world implications—from new approaches to wartime data analysis to tools for combating human trafficking. He talks to program manager Weishung Liu about the influences that led to the work and more. The post Called to serve: Tech, research, and positive impact appeared first on Microsoft Research .
Today, I’m talking with Illinois Governor JB Pritzker, who has a lot of thoughts about AI, tech, and privacy. If you’ve been listening to the…
Long-term memory enables language-model agents to personalize future interactions, but a finite memory bank cannot preserve every extracted fact. Existing memory gating policies are typically shared across users and therefore fail to account for user-specific patterns of future reuse. In this paper, we study memory gating at the fact level. We first propose a retrieval-based labeling strategy that labels each fact by whether it is later retrieved and incorporated into responses. We then formulate online memory gating as contextual apple tasting with delayed one-sided feedback, and propose Warm-Start UCB, which combines offline learning from existing data with online adaptation to the current user. Our theoretical analysis establishes a high-probability regret upper bound of {O}(Tεd{T}) when the delay window is at most O({T}), where T is the total number of rounds, ε is the initialization error, d is the dimension of the feature map, and {O}() hides constant and poly-logarithmic factors. The results of our experiment demonstrate that our method outperforms all baselines under different memory budgets and history lengths, establishing it as a promising personalized memory gating policy.
Nvidia's RTX Spark N1X processor is finally here, and we've been collectively testing a small fortune's worth of powerful laptops sporting this new "superchip." RTX…
Large language models (LLMs) can generate fluent outputs that lack source support or conflict with local operational constraints. We introduce Self-Reflective Chain-of-Thought (SRCoT), a modular architecture that organizes generation as a closed Retrieve--Reflect--Revise loop. SRCoT separates environment state, persistent knowledge, accumulated experience, and short-term memory into provenance-labeled stores. BM25 retrieval supplies local context; subsequent passes expose intermediate drafts, missing-information markers, follow-up queries, and optional claim--evidence mappings. These artifacts inform retrieval, revision, clarification, or abstention decisions. We examine the architecture through qualitative, trace-based case studies in two fully synthetic settings: emergency medical services and Munich-Sim, an emergency-operations simulation. Local prototype runs on an Apple M4-family laptop with 128 GB memory illustrate source-labeled documentation, explicit gap marking, and structured dispatch-style outputs. The traces also reveal provenance-assignment errors, incomplete evidence coverage, and unsafe assumptions that remain possible despite prompting constraints. These cases demonstrate report-level inspectability and local execution, not comparative superiority, hallucination reduction, clinical validity, or operational safety. The architecture separates retrieval and control from the model interface, but behavior across different models remains to be evaluated. SRCoT is a research proof of concept requiring human review, with its principal contribution lying in an inspectable control procedure rather than a new model.
There's a growing trend in the consumer tech world: gadgets designed to spy on everything and everyone around you, regardless of whether or not anyone…
The other day I whizzed past several people while running in the park, my playlist blasting at full volume. The next morning, I smiled at…
Ambi is releasing apps for iPhone, Mac, and Apple Watch that enables ambient capture
I'm embarking on a journey I've so far been hesitant to go on: actually using AI on the regular. Giving my most personal data to…
After launching nearly a month ago and spending several weeks as the top free app in Apple's App Store, the latest update to Meta's Muse…
Small language-model agents on edge devices must hold a persona and reason correctly at once, inside one context window that fills with conversational history and persona instructions. We study what happens to the logical part of such an agent when that history is long, misleading and persona-heavy (persona-logic interference), and present a Decoupling Architecture (AO-DA) that separates logical inference ("What") from persona expression ("How") into two inference paths on one INT4 base model with hot-swappable LoRA adapters. The logic path receives only the core turn and emits a verifiable structured state (Micro-State); the persona path renders it in character with the full history. In same-base-model ablations on an Apple M2 laptop (Llama-3.1-8B-Instruct and Gemma-3-4B-it, 4-bit; 480 runs over 4 pollution levels x 3 arms x 2 tasks x 2 personas x 5 seeds) we find: (i) the decoupled logic path is structurally invariant to pollution: its prompt stays at 180 (Llama) or 167 (Gemma) tokens while the mixed single-pass prompt grows from 242 to 1,203, and its outputs are byte-identical across levels (40/40); (ii) the mixed single pass degrades monotonically (composite logic score 0.669 to 0.150 on Llama, 0.487 to 0.150 on Gemma), mostly by failing to emit the required structured output (80-95% of runs on Llama, 100% on Gemma at the two highest levels); (iii) with the same pollution fed into the decoupled logic path, the dedicated-adapter, dedicated-format path is still more robust than the single pass on the 8B model (failure 0-20% vs 80-95%; paired Δ +0.30 to +0.50, Cliff's δ 0.50-0.85, Holm-adjusted p 0.03) but not on the 4B model, where both collapse. Separation costs one extra decode on a topic's first turn (28.2 s vs 18.2 s on Llama) and buys persona hot-swapping in 1.7 ms without re-running the logic path. Code, rubric, fixtures, adapters and logs are released.
