Tag
Explore by tags
Transforms research papers, natural-language specs, and technical descriptions into runnable code via a multi-agent system. Covers Paper2Code, Text2Web, and Text2Backend; scores 75.9% on OpenAI's PaperBench, ahead of top ML PhDs.
Synthesizes up to 90 minutes of multi-speaker speech in one pass, with as many as four voices in a single conversation. Pairs continuous acoustic and semantic tokenizers at a 7.5 Hz frame rate with a next-token diffusion head on an LLM backbone.
Compares standard human psychometric questionnaires (PVQ, BFI) with generation‑based profiling to test whether questionnaires predict real LLM responses. Finds big divergences: questionnaires exploit lexical cues and elicit alignment‑consistent answers, mischaracterizing LLM behavior on everyday queries.
Argues a single web-scale generative video model handles vision tasks zero-shot the way LLMs handle language. Probes Veo 3 on segmentation, edge detection, image editing, physical and affordance reasoning, and puzzles like maze solving and symmetry.
An open large language model pairing DeepSeek Sparse Attention (DSA) for cheaper long-context inference with a scaled RL pipeline. Authors claim parity with GPT-5, with a high-compute Speciale variant surpassing it and rivaling Gemini-3.0-Pro on reasoning.
Provides a plug-and-play inference engine that lets language models programmatically inspect, decompose, and recursively call themselves to handle very long contexts; supports local and cloud REPL sandboxes, multiple LLM backends, and trajectory logging/visualization.
Provides a conditional memory module that performs O(1) N‑gram lookups and fuses static embeddings into transformer hidden states — enables offloading large embedding tables to host memory with minimal inference overhead.
Reduces object-driven shortcut learning in zero-shot compositional action recognition by enforcing temporal verb cues and regularizing against frequent object-verb co-occurrence priors. Proposes RCORE with Co-occurrence Prior Regularization (treats frequent co-occurrences as hard negatives) and Temporal Order Regularization. Evaluated on Sth-com and EK100-com with improved compositional generalization.
Lightweight, Markdown-only skill pack that lets LLM agents autonomously run ML research workflows—literature survey, idea discovery, cross-model review loops, experiment automation and paper writing—designed for Claude Code, Codex CLI, Cursor and local model setups.
Runs an autonomous self-improvement loop where a meta agent crafts a task-specific agent, a target agent executes trials, and a feedback agent updates both harness (code) and model weights—provider-agnostic profiles with reproducible runs and a live dashboard.
Converts text to natural-sounding speech across 600+ languages in a zero-shot way, with short-reference voice cloning and fine-grained voice-design controls; uses a diffusion language-model-style architecture to balance quality and very low inference latency.
Provides a diagnostic suite that audits video-understanding benchmarks to find samples solvable without visual or temporal input, filters those shortcuts, and produces a distilled video-native testbed that reveals major capability gaps in current Video-LLMs.