Dự án huấn luyện AI lớn đang bắt đầu tích hợp dữ liệu suy luận kiểu Chain-of-Thought (CoT) vào tập luyện. Điều này giúp mô hình hiểu sâu hơn các bước logic, nâng cao khả năng giải quyết vấn đề. Nguồn: [Reddit]([link]) #AINN #HocMay #CoT #KhoaHocDuLieu #AIResearch #MachineLearning #NeuralNetworks #TuitionLearning

https://www.reddit.com/r/LocalLLaMA/comments/1przir5/big_training_projects_appear_to_be_including_cot/

chatGPTのプロンプト作成の基礎テクニック9選。賢いプロンプトの作り方

https://www.aiandemily.com/chatgpt%e3%81%ae%e3%83%97%e3%83%ad%e3%83%b3%e3%83%97%e3%83%88%e4%bd%9c%e6%88%90%e3%81%ae%e5%9f%ba%e7%a4%8e%e3%83%86%e3%82%af%e3%83%8b%e3%83%83%e3%82%af9%e9%81%b8%e3%80%82%e8%b3%a2%e3%81%84%e3%83%97/?feed_id=3163&_unique_id=693a5cbad8718&Mastodon

#chatgpt #chatgptコツ #chatgpt活用方法 #AI部 #chatgptプロンプト #chatgpt使い方 #chatgptプロンプトエンジニアリング #chatgptプロンプトエンジニア #promptengineeringguide #CoT #AI #プログラムコード生成 #プログラムコード生成 #chatgpt #chatgptコツ #c...

Cột đèn chiếu sáng sân vườn là giải pháp chiếu sáng và trang trí không gian ngoại thất như sân vườn, công viên, lối đi, khu nghỉ dưỡng. Sản phẩm có thiết kế tinh tế, chiều cao thấp từ 2–4m, đa dạng mẫu mã, chất liệu bền bỉ.

Xem ngay: https://congtydenled.com.vn/cot-den-chieu-sang/san-vuon
Liên Hệ:
Hotline 0332599699
Địa chỉ: Số 3D2, KDT Cầu Diễn, Bắc Từ Liêm, Hà Nội
Website: https://congtydenled.com.vn/
#haledstore
#công ty đèn LED
#cot-den-chieu-sang
#cột đèn chiếu sáng
#cotdenchieusangsanvuon
#cot-den-chieu-sang-san-vuon

"Phương pháp LEASH (Logit-Entropy Adaptive Stopping Heuristic) giúp tối ưu hóa quá trình suy luận Chain-of-Thought (CoT) trong các mô hình ngôn ngữ. Thay vì tạo giải thích dài, LEASH dừng lại khi xác suất token và cải thiện top-logit ngừng thay đổi, tiết kiệm 30-35% token và 27% thời gian. Tuy nhiên, độ chính xác giảm ~10%. Phù hợp với mô hình được tinh chỉnh. #AI #ML #NLP #HiệuSuất #HàngĐầu #MachineLearning #SuyLuận #CoT"

https://www.reddit.com/r/singularity/comments/1orvj0h/logitentropy_adapt

📸 Aí está parte do COT – Core Organizing Team do GLAM Wiki 2025, reunido em Lisboa para preparar todos os detalhes deste grande encontro internacional! 🌍

De 30 de outubro a 1 de novembro com:
🎤 Painéis e apresentações
🛠️ Oficinas práticas
💻 Hackathon
🧭 Sessões de estratégia
🏛️ Tours culturais

👉 Nota: as inscrições terminam a 30 de setembro.

#GLAMWiki2025 #Lisboa #COT #GLAM #Wikimedia #WikiLovers #WikimediaPortugal #WMPT #OpenKnowledge #OpenCulture

https://aipapersacademy.com/hierarchical-reasoning-model/

#ai #machinelearning #hrm #cot

[Перевод] LLM и их хрупкая логика: новое исследование ставит под сомнение Chain-of-Thought

Новое исследование учёных из Университета штата Аризона показывает: знаменитое «цепочечное рассуждение» (Chain-of-Thought, CoT) в больших языковых моделях (LLM) скорее похоже на «хрупкий мираж», чем на проявление подлинного интеллекта. Эта работа продолжает традицию критического анализа глубины рассуждений LLM, но в отличие от предыдущих исследований предлагает уникальный взгляд через призму «распределения данных», который позволяет понять, где и почему CoT систематически даёт сбой.

https://habr.com/ru/companies/technokratos/articles/939072/

#большие_языковые_модели #искусственный_интеллект #ai #llm #cot #chain_of_thoughts

How can #AI "reasoning" models be more efficient?

Liu trained them on only chains-of-thought (CoTs) — no prompts.

This seemed to teach models
- #CoT reasoning
- conditions that trigger longer CoTs
- default to shorter CoTs in other conditions

🔓https://doi.org/10.48550/arXiv.2506.12860

#CogSci

https://www.wacoca.com/news/2617769/ インド、綿花の輸入関税を9月末まで一時停止 | ロイター #AGRI #AMERS #APPA #Asia #ASXPAC #BD #BMAT #CDTY #com. #COT #CYCP #CYCS #DEST:NOJPWDM #dip #EMRG #forest #Gen #in #India #JFOR #JLN #MINE #MTAL #MTPIX #NAMER #POL #POTUS #PXP #SASIA #SFTS #TEX #TOPCMB #TOPNWS #TRD #TRF #TRN #TXTL #US #wash #wear #インド

https://www.walknews.com/1018389/ インド、綿花の輸入関税を9月末まで一時停止 | ロイター #AGRI #AMERS #APPA #ASIA #ASXPAC #BD #BMAT #CDTY #COM #COT #CYCP #CYCS #DEST:NOJPWDM #DIP #EMRG #FOREST #GEN #IN #India #JFOR #JLN #MINE #MTAL #MTPIX #NAMER #POL #POTUS #PXP #SASIA #SFTS #TEX #TOPCMB #TOPNWS #TRD #TRF #TRN #TXTL #US #WASH #WEAR #インド

Just in case you were holding your breath for LLMs that had "reasoning" or "chain of thought" to let them "think" more logically, you might want to consider how the AI companies are going about this.

Which path do you think they've pursued?
(a) link the power of LLM natural language skills with the existing non-generative AI problem-solving, path-exploring, optimization tools, or
(b) get their LLMs to produce a set of steps that statistically looks like it would be a chain of thought, interpolating the already collected training material -- an approach that breaks as soon as the problem to be broken down can't be recognized in the training data?

https://www.youtube.com/watch?v=TSaiuXPe6cw

Asking a chatbot to explain its reasoning (it can't, it just sounds like it) or explain its errors (it can't, it just sounds like it) will produce convincing but completely fabricated text. Asking a chatbot to break a problem down into a series of logical steps and then solve them will only work when there are close enough examples of very similar problems in its training. It can adopt the form and the prosaic features in its answers, but it can't (yet) apply logic.

Maybe some time in some unlimited-power future, the depth of learning in a huge model will abstract and embed the structure of logic, reasoning, analogy, induction, probability etc., from some body of human writing, but it appears (to me) almost impossible to extract that from any existing corpus. The currently available training material has a wealth of examples of simple logic and way too many examples of bad reasoning.

If you know of researchers that are following path (a) above with any success, let me know. I think that's the only way we'll get any useful explorative, extrapolative, less-brittle AI.

#AI #CoT #LLM

#Youtube #Video: #AI ‘chain of thought’: still just a ‘brittle mirage’
https://www.youtube.com/watch?v=TSaiuXPe6cw
#LLM #reasoning #COT

Die #KI befindet sich mit seiner #Denkfähigkeit per #CoT also auf der höhe der meisten öffentlichen Diskurse:
"Dieses Ergebnis sei sprachlich plausibel, aber logisch inkonsistent."
https://the-decoder.de/naechste-studie-aeussert-zweifel-an-denkfaehigkeit-von-reasoning-modellen/?utm_source=firefox-newtab-de-de

Is #chainofthought #Reasoning of #LLMs a Mirage?

"... Our results reveal that #CoT reasoning is a brittle mirage that vanishes when it is pushed beyond training distributions. This work offers a deeper understanding of why and when CoT reasoning fails, emphasizing the ongoing challenge of achieving genuine and generalizable reasoning.

... Our findings reveal that CoT reasoning works effectively when applied to in-distribution or near
in-distribution data but becomes fragile and prone to failure even under moderate distribution shifts.
In some cases, LLMs generate fluent yet logically inconsistent reasoning steps. The results suggest that what appears to be structured reasoning can be a mirage, emerging from memorized or interpolated patterns in the training data rather than logical inference.

... Together, these findings suggest that LLMs are not principled reasoners but rather sophisticated simulators of reasoning-like text."

Do #AI models perform better as they do more chain-of-thought (#CoT) reasoning?

When is more reasoning no longer worth it?

This paper finds near-optimal results when CoT reasoning terminates ... almost immediately?

More reason to think CoT's overrated?

https://doi.org/10.48550/arXiv.2505.15684

Anthropic researchers discover the weird AI problem: Why thinking longer makes models dumber
https://zurl.co/IeLIu
#ai #agenticai #cot

Research leaders urge tech industry to monitor AI’s ‘thoughts’

A key feature of AI reasoning models are their chains-of-thought or CoTs — an externalized process in which AI models work through problems, similar to how humans use a scratch pad to work through a difficult math question. Reasoning models are a core technology for powering AI agents

#artificialintelligence #CoT #AI #technology #tech #anthropomorphism

https://techcrunch.com/2025/07/15/research-leaders-urge-tech-industry-to-monitor-ais-thoughts/

#AI systems that reason in natural language can expose their intentions, making Chain-of-Thought (#CoT) monitoring a promising safety lever

🔗 https://arxiv.org/abs/2507.11473

#ResponsibleAI

When many of the most influential AI researchers agree on something, it’s worth paying attention - like this paper on Chain-of-Thought monitorability: a fragile but vital path to AI safety. Imagine spotting misbehaviour before it happens. Transparency matters. #AI #ChainOfThought #CoT #AISafety

Chain of Thought Monitorabilit...

Mastering Reasoning with Code: Chain of Thought Pipelines in Java
Build smarter, traceable AI workflows using LangChain4j, Quarkus, and structured step-by-step logic
https://myfear.substack.com/p/chain-of-thought-java-langchain4j-quarkus
#Java #Quarkus #LangChain4j #CoT #Reasoning