Search posts, papers, and topics
ParparVM’s HashMap suffered catastrophic miss latency due to linear probing on dense integer keys. By adopting CPython‑style perturbed probing (Swiss‑table style) and extending tagged immediate values to more primitives, miss latency dropped from 32 s to ~45 ms, allocation pressure fell dramatically, and overall performance stayed roughly flat despite a modest hit‑time slowdown.
E2A‑Bench is a 969‑query benchmark for financial chart reasoning that evaluates vision‑language models across a full evidence‑to‑action chain using four metrics (UCR, RCI, ECI, NDR). Experiments on 20 VLMs expose hidden failures: low‑UCR models have only 6.4 % directional coverage, oracle‑aided verification cuts unsupported claims but can kill coverage, and fine‑tuning inflates BUY:SELL ratios by…
Lightning Weave is a post‑training framework that composes specialist capabilities from separate models into one student via on‑policy distillation, improving both accuracy and token efficiency on math and code benchmarks.
OmniHarness introduces a symbolic‑policy framework that extracts reusable procedural knowledge from multimodal LLM‑driven visual generation runs. By decoupling task logic from instance inputs, the system can instantiate, adapt, and compose policies for new visual tasks, using intermediate verification for on‑the‑fly refinement while keeping the underlying model frozen. Self‑directed practice task…
The paper experiments with prompting six frontier LLM families using a school‑audience framing and finds their architectural design suggestions converge on a common pattern, while removing the framing yields diverse outputs. It highlights a possible shared design prior among models and introduces the term "epistemic jailbreak" for the loss of provenance in speculative answers.
This paper introduces SpectralShift, a spectral reparameterization method for extending the context window of Gated DeltaNet (GDN) linear attention models. It reconfigures the decay spectrum by enhancing slow propagation and preserving fast-decaying modes, consistently improving long-context capabilities during continual pretraining.
The author reverse‑engineers Factorio’s RNG, confirms it’s the Boost taus88 generator (a 3‑LFSR XOR combo), proves equivalence with the game’s compiled code via decompilation and symbolic Python checks, and explains how the weak LFSR structure lets you predict future “random” quality outcomes.
Realtime-Venus is a full‑duplex, multimodal dialogue system built from two separately trained 9B models (Omni for audio‑visual, Audio for spoken interaction). It uses a shared causal timeline and a dual‑loop runtime that lets foreground interaction continue while a background Harness executes delegated tasks asynchronously. The paper reports benchmark scores where Realtime‑Venus‑Omni leads on six…
ParparVM’s GC was tuned by lowering the allocation‑trigger floor, adding configurable thresholds, parallel marking, mutator assistance, and proper weak/soft reference handling. These changes cut RSS from 98 MB to 38 MB, reduced worst‑case GC pauses from seconds to sub‑second, and improved cache hit rates with a recency‑based eviction policy.
StepAudio 3 Realtime is an audio‑language foundation model that runs a continuous listen‑converse‑think‑act loop. It introduces Deep Perception for rich acoustic cue extraction, Seamless Duplex for handling pauses/back‑channels, and a Think‑While‑Speaking mechanism that lets the model reason in parallel with speech output. On benchmarks it scores 73.0 macro avg on StepAudioChat, 90.6 on MMSU, 98.…
The paper reframes fine‑tuning of instruction‑tuned LLMs as a direction‑selection problem under a fixed behavioral‑drift budget, showing that the update direction, not magnitude, determines trade‑offs between target performance and capability preservation. In QA‑only fine‑tuning of Qwen‑3 models, layer‑selective probing finds effective directions that boost scientific reasoning and multilingual t…
The paper builds compositional rule‑induction task families and generates isomorphic variants to probe systematicity in reasoning models. Experiments reveal that models frequently fail on equivalent variants despite solving the original task, indicating a lack of systematic reasoning.
The paper shows that when specialist LLMs are trained only on QA pairs (no explicit reasoning supervision), their optimization implicitly selects a latent distribution of reasoning trajectories. By treating the distilled student as an agnostic probe—since it inherits only the sampled trajectories—the authors empirically demonstrate a strong correlation (across 27 specialist‑student pairs) between…
The paper proposes the Convergent Emergence Hypothesis that few‑shot in‑context learning (ICL) shares a common difficulty profile across domains. Using a unified task suite, the authors evaluate ICL on six modalities—language, genome, integer sequences, time‑series, images, and proteins—showing that paired‑mapping ICL emerges in all and that per‑task benefits correlate across five modalities, sup…
Firefox 154 shipped with GeForce NOW support and a PiP playback‑speed control; the team moved to a 2‑week release cadence, added a Linux‑system‑theme option in Nova, and landed dozens of bug fixes across Add‑ons, WebExtensions, DevTools, WebDriver and the New Tab page, many driven by community contributors.
Safari Technology Preview 252 adds several WebKit platform updates, including new CSS features like named-feature() in @supports and unprefixed user-select, plus bug fixes across accessibility, animations, HTML, and JavaScript. The release also corrects worker termination, Promise.try behavior, and line‑clamp rendering, which can affect Safari‑targeted web apps.
A practical guide to refactoring a monolithic onboarding API into an event‑driven architecture using ASP.NET Core, AWS SNS/SQS (or EventBridge), the outbox pattern, and idempotent consumers, with concrete code snippets and operational tips.
The Arduino VENTUNO Q board combines a Linux‑capable MPU with a real‑time MCU, enabling a single platform for ROS 2 navigation, sensor fusion, and low‑latency motor control in autonomous mobile robots. The post shows how LiDAR, cameras, edge‑AI and an MPU‑to‑MCU communication layer let developers prototype AMRs step by step.
Hostinger built an internal 174‑question Mindset Map assessment to surface candidates' and employees' motivations, learning styles, and adaptability. The AI‑processed results are used as one input for hiring and career conversations, not as a decision‑making metric.
StepAudio 3 Gen is a general‑purpose audio generation model that replaces diffusion with a discrete autoregressive generator over residual vector quantization tokens. Using a 16‑layer RVQ tokenizer and progressive pretraining, it reaches state‑of‑the‑art zero‑shot TTS and voice‑design performance while handling speech, vocals, sound effects, and music.
The paper introduces Generalized Agent Iteration (GAI), a formal framework that unifies classical iterative policy improvement (GPI) and recursive self‑improvement (RSI). GAI treats an agent as a set of modifiable components and models learning as a loop of evaluation and improvement. Two binary “dials”—whether the improvement mechanism is internal to the agent and whether the evaluation standard…
Codename One engineers dissect why benchmark numbers can be misleading, then share concrete work on GC tuning, proper weak/soft references, and a new probing sequence for their open‑addressed HashMap that cuts miss‑probe counts from >16 k to ~1.5 per lookup.
RelateAnything is a 53 M‑parameter model that predicts arbitrary textual relations between any supplied image regions, using a text‑embedding bank instead of a fixed classifier. Trained on a new 4.3 M‑relation dataset, it outperforms prior open‑vocab methods by 2.3‑3.5× mean recall while running in ~20 ms per frame.
Moving the Firefox build output directory (MOZ_OBJDIR) out of the source tree speeds up IDE indexing by removing massive artifact trees. The downside is that generated code shows up as missing symbols, so developers must balance indexing performance against code navigation.
The paper presents SteerDuplex, a full‑duplex speech dialogue model that can be steered along tone, persona, and speed via instruction following, and introduces the SteerBench benchmark to evaluate such steerability. Supervised training yields a 44.5 % pass‑rate lift, and reinforcement‑learning fine‑tuning improves interruption handling and reduces pause barge‑ins, though reward hacking remains a…
A step‑by‑step tutorial showing how to build a Twilio voice agent in PHP (using OpenSwoole) that calls OpenAI’s gpt‑4o‑mini model and can invoke an external REST API (Cat Facts) via tool‑calling, wiring everything together with Twilio Conversation Relay and a WebSocket server.
Latent Interface Training (LIT) first learns a goal‑conditioned action prior without visual input, then adds a pose‑supervised latent interface as the only visual conditioning path. Applied to several vision‑language‑action models, LIT cuts vision‑action shortcuts and lifts LIBERO‑Plus success by 3.9–10.7 points and real‑world task success by 13.3–16.7 points under distribution shifts.
