What turned up in the last three days — collected by machine, published by hand.
What gets picked here ends up in Teardowns →
Collected 2026-09-08 23:01 UTC · 46 kept from 280
Source HF Daily Papers124 upvotes3 comments
13 authorsno limits stated
“Command-line coding agents (e.g., Claude Code, Gemini CLI) can already read and write files and sustain long sessions, yet end-…”
Published 2026-08-30
Fetched 2026-09-07 23:01 UTC · 17101143ec4ae6bc
Torn from — the abstract
Thread ↗
Source HF Daily Papers73 upvotes1 comments
FlowBalance4 authorsno limits stated
“FlowBalance improves reasoning models via verifier-calibrated self-guidance using trajectory-level score reweighting and profil…”
They claim — On mathematical reasoning, FlowBalance improves average performance over FlowRL on both Qwen3-4B and Qwen3-8B, while also improving training speed and stability, avoiding direct OPSD's re…
Published 2026-09-02
Fetched 2026-09-08 16:21 UTC · 437efcb18b365a17
Torn from — the abstract
Thread ↗
Source arXiv17 upvotes2 comments
cs.AI9 authorsno limits stated
“Layer dropout (a.k.a. stochastic depth) has been shown to enable faster training, higher accuracy, and robustness to zero-shot …”
They claim — For a given number of training steps, LLMs can achieve lower or similar validation loss while saving upto 25% of training FLOPs.
Published 2026-09-04
Fetched 2026-09-07 01:32 UTC · 60c554528e9aa388
Torn from — the abstract
Thread ↗
Source arXiv17 upvotes3 comments
math.OC6 authorsno limits stated
“Large language models (LLMs) are increasingly used to formulate optimization models from natural-language problem descriptions,…”
Published 2026-09-04
Fetched 2026-09-07 01:32 UTC · 1ba5a6f5e64b9cda
Torn from — the abstract
Thread ↗
Source HF Daily Papers33 upvotes2 comments
text-promptable segmentation5 authorsstates limits
“ENEAS unifies text-prompted instance tracking and open-concept semantic discovery via temporal memory extension and a verificat…”
Published 2026-09-02
Fetched 2026-09-08 16:21 UTC · a5384a599ed5599b
Torn from — the abstract
Thread ↗
Source HF Daily Papers21 upvotes2 comments
Open Information Extraction8 authorsno limits stated
“Enoki is an open information extraction framework that unifies claim-level verification and span-level hallucination localizati…”
Published 2026-08-31
Fetched 2026-09-07 23:01 UTC · ab91839c3bae8764
Torn from — the abstract
Thread ↗
Source HF Daily Papers29 upvotes3 comments
audio-video diffusion models7 authorsno limits stated
“Audio-video diffusion models exhibit bidirectional semantic leakage through cross-modal attention pathways, which can be diagno…”
Published 2026-09-02
Fetched 2026-09-07 23:01 UTC · 496e910644d364a4
Torn from — the abstract
Thread ↗
Source HF Daily Papers13 upvotes2 comments
reasoning operations6 authorsno limits stated
“Distinct reasoning operations in language models are geometrically separable in hidden representations, with structure emerging…”
Published 2026-09-03
Fetched 2026-09-07 23:01 UTC · 8fd8b68e8d2fd445
Torn from — the abstract
Thread ↗
Source HF Daily Papers53 upvotes4 comments
search agents9 authorsno limits stated
“Two large-scale search agents are trained via a multi-stage pipeline combining supervised fine-tuning and reinforcement learnin…”
Published 2026-09-02
Fetched 2026-09-07 23:01 UTC · bf933da181c11d7a
Torn from — the abstract
Thread ↗
Source HF Daily Papers38 upvotes2 comments
spoken dialogue model4 authorsno limits stated
“Motion-Omni is an end-to-end framework that jointly generates spoken dialogue and full-body co-speech motion from shared hidden…”
They claim — We further release SwDA-500 and, to our knowledge, the first public evaluation protocol for stochastic open-ended full-body spoken dialogue, matching audio across motion systems while uni…
Published 2026-08-27
Fetched 2026-09-07 23:01 UTC · 1ae2e3cb2c0f6040
Torn from — the abstract
Thread ↗
Source HF Daily Papers13 upvotes2 comments
topology-aware diffusion transformer7 authorsno limits stated
“UniMate is a unified diffusion transformer that generates articulated motion for arbitrary skeletons from text and rigged 3D as…”
They claim — UniMate introduces a topology-aware diffusion transformer, which integrates skeletal topology into attention via three mechanisms: (1) a graph-aware attention bias from pairwise joint rel…
Published 2026-09-03
Fetched 2026-09-07 23:01 UTC · f7aead1ac8cf4d5a
Torn from — the abstract
Thread ↗
Source HF Daily Papers21 upvotes1 comments
compositional 3D representation12 authorsno limits stated
“Adapting a single-object 3D generative prior to multi-view observations enables scalable compositional mesh reconstruction of d…”
Published 2026-09-03
Fetched 2026-09-07 23:01 UTC · acfc206edb29736d
Torn from — the abstract
Thread ↗Nothing on this page matches what you picked.
Papers, repos, models, blogs, Reddit and podcasts — once a day, by machine. Everything collected gets taken apart once — the README, the abstract, the model card — and what that turns up is kept for good, whether or not it becomes a card. Titles and links go to the original; nothing is reproduced here.
The grey line on each card is what the teardown found, not what the source advertises: the licence, the last commit, the model it was built on, whether the paper states its own limits. Open “What we saw” to check what was read to get it. Where it says listing only, the original could not be read — some sites refuse us, and we do not pretend otherwise.
A line in quotation marks is a sentence lifted from the source itself — their number, not a measurement of ours. We copy it and leave it unchecked; that is the point of quoting it.
Topics are assigned by matching words in the original title, quote and tags against a fixed list — no model reads the card. Some cards match nothing and carry no topic at all; pick a topic and those drop out. Source and topic narrow together.
Each card keeps the sentence we actually read at the moment we read it, plus a fingerprint of the response we received. That sentence — not the whole page — is what we can still stand behind if the original later changes or disappears.