TERRA: Terrain-Aware Reconstruction, Retargeting and Control for Musculoskeletal Locomotion
Paper โข 2609.38653 โข Published โข 39
AI & ML interests
None defined yet.
Recent Activity
View all activity
chengkunliย
submitted a
paper to Daily Papers 9 days ago
omarkamaliย
posted an update about 1 month ago
Post
183
Apparently grammar can travel without a passport ๐
We trained a new neural grammar model *Tarakib* across 166 languages on top of Sawtone [1] - itself trained on nearly 400 langs. We then tested it on 28 languages that contributed no training updates to the grammar task.
The surprising result: it still learned useful parts-of-speech structure based on the foundational linguistic representation offered by Sawtone.
Performance varied considerably, and the transfer was not limited to close, same-script relatives but crossed writing systems. We found connections across language families, and about a quarter of the stable transfer links crossed writing systems.
This does not mean universal parsing is solved. But it suggests that Sawtone's representations contain universal grammatical clues that a shared model can reuse in languages it was never directly taught to parse.
A small step toward "spaCy for every language" โ๏ธ
1. https://www.omneitylabs.com/models/sawtone
We trained a new neural grammar model *Tarakib* across 166 languages on top of Sawtone [1] - itself trained on nearly 400 langs. We then tested it on 28 languages that contributed no training updates to the grammar task.
The surprising result: it still learned useful parts-of-speech structure based on the foundational linguistic representation offered by Sawtone.
Performance varied considerably, and the transfer was not limited to close, same-script relatives but crossed writing systems. We found connections across language families, and about a quarter of the stable transfer links crossed writing systems.
This does not mean universal parsing is solved. But it suggests that Sawtone's representations contain universal grammatical clues that a shared model can reuse in languages it was never directly taught to parse.
A small step toward "spaCy for every language" โ๏ธ
1. https://www.omneitylabs.com/models/sawtone
[Join request] Verification and provenance for AI agents
2
#29 opened about 1 month ago
by 
AiOpsEnabler
marcusintheskyย
authored 2
papers about 1 month ago
Portfolio Risk Bounds without Cross-Asset Return Covariances: Distributional Fields from Language-Model Representations
Paper โข 2608.29692 โข Published โข 2
Wasserstein-Barycentric Interaction Fields for Spatial Factor Models: Evidence from Language-Model Representations
Paper โข 2608.29669 โข Published โข 2
marcusintheskyย
submitted 2
papers to Daily Papers about 1 month ago
Portfolio Risk Bounds without Cross-Asset Return Covariances: Distributional Fields from Language-Model Representations
Paper โข 2608.29692 โข Published โข 2
Wasserstein-Barycentric Interaction Fields for Spatial Factor Models: Evidence from Language-Model Representations
Paper โข 2608.29669 โข Published โข 2
[Support] Community Articles
๐ฅ๐ 2
115
#5 opened over 2 years ago
by 
victor
Join request pending for a week, anything else needed?
1
#28 opened 3 months ago
by 
yuncheol-ncdtech
[Support] Community Articles
๐ค๐ฅ 2
115
#5 opened over 2 years ago
by victor
AmanPriyanshuย
authored 3
papers 2 months ago
Antares: Foundation Models for Agentic Vulnerability Localization
Paper โข 2608.02407 โข Published โข 3
FAPO: Fully Autonomous Prompt Optimization of Multi-Step LLM Pipelines
Paper โข 2606.19605 โข Published โข 11
Got a Secret? LLM Agents Can't Keep It: Evaluating Privacy in Multi-Agent Systems
Paper โข 2605.27766 โข Published โข 1
Post
131
๐ฅฌ LettuceDetect v2 โ span-level hallucination detection for code, tool output, and structured documents.
Hallucination detectors are trained on document QA, but agents ground their answers in source code, tool output and markdown. On code-agent answers, existing detectors reach 0.17 span-F1 and even 550B zero-shot judges at most 0.22.
We built a unified span-level benchmark โ 74,285 newly constructed examples (145K+ with RAGTruth and 14-language PsiloQA folded in), every span typed and character-labeled โ and trained two detectors on it:
๐ค KRLabsOrg/lettucedect-v2-qwen-2b โ generative, typed spans + explanations in one pass, 32K context, **0.689 span-F1** (0.60 on code-agent)
โก KRLabsOrg/lettucedect-v2-mmbert-base โ 307M multilingual encoder for high-throughput setups
๐ท๏ธ KRLabsOrg/lettucedect-v2-taxonomy-head โ types the spans of any binary detector
It also reaches the best reported English PsiloQA IoU (0.724) and 81.8 RAGTruth example-F1, so specializing on code didn't cost general RAG performance.
๐ Dataset: KRLabsOrg/lettucedetect-code-hallucination
๐ Paper: https://arxiv.org/abs/2607.00895
The models are now integrated natively into vLLM Semantic Router โ joint blog post on how it works: https://vllm-sr.ai/blog/lettucedetect-v2-generative-hallucination-detection
Hallucination detectors are trained on document QA, but agents ground their answers in source code, tool output and markdown. On code-agent answers, existing detectors reach 0.17 span-F1 and even 550B zero-shot judges at most 0.22.
We built a unified span-level benchmark โ 74,285 newly constructed examples (145K+ with RAGTruth and 14-language PsiloQA folded in), every span typed and character-labeled โ and trained two detectors on it:
๐ค KRLabsOrg/lettucedect-v2-qwen-2b โ generative, typed spans + explanations in one pass, 32K context, **0.689 span-F1** (0.60 on code-agent)
โก KRLabsOrg/lettucedect-v2-mmbert-base โ 307M multilingual encoder for high-throughput setups
๐ท๏ธ KRLabsOrg/lettucedect-v2-taxonomy-head โ types the spans of any binary detector
It also reaches the best reported English PsiloQA IoU (0.724) and 81.8 RAGTruth example-F1, so specializing on code didn't cost general RAG performance.
๐ Dataset: KRLabsOrg/lettucedetect-code-hallucination
๐ Paper: https://arxiv.org/abs/2607.00895
The models are now integrated natively into vLLM Semantic Router โ joint blog post on how it works: https://vllm-sr.ai/blog/lettucedetect-v2-generative-hallucination-detection
- 3 replies
[Support] Community Articles
๐ค๐ 2
115
#5 opened over 2 years ago
by victor
[Support] Community Articles
๐๐ค 2
115
#5 opened over 2 years ago
by victor
chengkunliย
authored a
paper 3 months ago
chengkunliย
submitted a
paper to Daily Papers 3 months ago
adaamkoย
submitted a
paper to Daily Papers 3 months ago
Post
1631
Uhh did Opus 4.8 cheat on PostTrainBench??
it found an API key in the PostTrainBench environment that allowed it to generate synthetic training data without using GPU hours, boosting the base model by 0.4913
Source: https://posttrainbench.com/traces/run.html?id=claude_non_api_max_claude-opus-4-8_10h_run1__healthbench_Qwen_Qwen3-4B-Base_17315102#tab=trace
it found an API key in the PostTrainBench environment that allowed it to generate synthetic training data without using GPU hours, boosting the base model by 0.4913
Source: https://posttrainbench.com/traces/run.html?id=claude_non_api_max_claude-opus-4-8_10h_run1__healthbench_Qwen_Qwen3-4B-Base_17315102#tab=trace
- 1 reply