Outerport

60 posts

Outerport banner
Outerport

Outerport

@outerport

Accelerating physical product development with AI agents that can understand engineering documents and CAD for DFM & process & supply chain simulations (YC S24)

Katılım Temmuz 2024
4 Takip Edilen726 Takipçiler
Outerport retweetledi
Towaki Takikawa / 瀧川永遠希
Towaki Takikawa / 瀧川永遠希@yongyuanxi·
ダイキン工業様との協業を発表しました！設計でのAI活用において、必要なノウハウが詰まっている図形・図表を構造化するOuterportの技術を本格導入していただきました。 CG/CV/AIの技術を総動員して、新たなハードウェア技術の設計・開発・製造がもっと速くできる世界を目指して精進いたします🛠️🏭⚙️
Towaki Takikawa / 瀧川永遠希 tweet media
日本語
12
19
191
39.3K
Outerport retweetledi
Towaki Takikawa / 瀧川永遠希
Towaki Takikawa / 瀧川永遠希@yongyuanxi·
Parsing CAD drawings is one thing but parsing architectural plans takes the craziness up a notch
Towaki Takikawa / 瀧川永遠希 tweet media
English
6
17
184
15K
Outerport retweetledi
Towaki Takikawa / 瀧川永遠希
Towaki Takikawa / 瀧川永遠希@yongyuanxi·
Yeah document parsing is cool, but what about CAD drawing parsing... (but with documents too 🥺)
Towaki Takikawa / 瀧川永遠希 tweet media
English
34
45
609
94K
Outerport
Outerport@outerport·
MetaCLIP 2: A Worldwide Scaling Recipe
Outerport tweet media
English
1
0
6
1.3K
Outerport
Outerport@outerport·
Taming the Untamed: Graph-Based Knowledge Retrieval and Reasoning for MLLMs to Conquer the Unknown Bowen Wang, Zhouqiang Jiang, Yasuaki Susumu, Shotaro Miwa, Tianwei Chen, Yuta Nakashima
Outerport tweet media
English
0
0
1
677
Outerport
Outerport@outerport·
Can LLMs Replace Humans During Code Chunking? The MITRE Corporation Benchmarked on legacy government code written in ALC, MUMPS, Assembly.... LLMs 20% more factual and 10% more useful than human partitioning
Outerport tweet media
English
1
0
2
446
Outerport
Outerport@outerport·
Controlled Retrieval-augmented Context Evaluation for Long-form RAG Jia-Huei Ju, Suzan Verberne, Maarten de Rijke, Andrew Yates arxiv.org/abs/2506.20051
Outerport tweet mediaOuterport tweet media
Română
0
0
1
280
Outerport
Outerport@outerport·
Seeing is Believing? Mitigating OCR Hallucinations in Multimodal Large Language Models Zhentao He, Can Zhang, Ziheng Wu, Zhenghao Chen, Yufei Zhan, Yifan Li, Zhao Zhang, Xian Wang, Minghui Qiu
Outerport tweet media
Filipino
0
0
2
474
Outerport
Outerport@outerport·
Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding Xiaoyi Zhang, Zhaoyang Jia, Zongyu Guo, Jiahao Li, Bin Li, Houqiang Li, Yan Lu
Outerport tweet mediaOuterport tweet media
Filipino
1
0
2
866
Outerport
Outerport@outerport·
When you drag select and it selects multiple columns of text at once 😔💪
Outerport tweet media
English
0
0
4
455
Outerport
Outerport@outerport·
The Viability of Crowdsourcing for RAG Evaluation (SIGIR 2025) Lukas Gienapp, Tim Hagen, Maik Fröbe, Matthias Hagen, Benno Stein, Martin Potthast, Harrisen Scells
Outerport tweet mediaOuterport tweet media
Deutsch
1
0
2
408
Outerport
Outerport@outerport·
@FJ000RD Makes sense! Empirically we've also found that a simple trick of chunking the docs into smaller (say 10k) segments works wonders for accuracy. Also I guess looking at systems like o3 a lot of "RAG-ish" behavior will be implicit in the agent's behavior
Outerport tweet media
English
1
0
1
28
Ford Lascari
Ford Lascari@FJ000RD·
I think largely not really. It just wont be worth the couple percentage points of accuracy/ quality you will get if say you are dropping 100-200 pages of docs (60-120k tokens). Which i feel most people / use cases will find just dumping it in the context window works good enough.
English
1
0
1
11
Ford Lascari
Ford Lascari@FJ000RD·
This is why RAG will never die, especially if you're aiming for the top 1%. Context is what keeps you ahead of those relying on generic data. RAG pulls in high-value tokens, thus positioning the LLM more effectively in latent space, optimizing task completion. This will always be critical.
EP@eptwts

everyone can now access all the information on the internet with 1 prompt... but that doesn't make all information valuable knowing where to get information from & how to qualify it is gonna be a great skill to have going forward i think we're heading into a world where selling access to specialized knowledge-bases that users can retrieve data from will be common a new type of infoproduct in the AI era the HQ knowledge-base tailored to your use-case will undoubtedly trump the massive info-dump found in an LLM's training data

English
2
0
3
317

Keşfet

@FJ000RD @elonmusk @BarackObama @taylorswift13 @cristiano @BillGates @NASA @nikifrancismediavine