The #OpenAI paper by Baker et al, "Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation" comes to a troubling conclusion: #LLM s with #reasoning or #ChainOfThought (#CoT) capabilities might learn to obfuscate their own CoT from human users if they are being penalized for displaying "wrong" (i.e. reward hacking or misalignment) reasoning.

As a result, OpenAI strongly advises against applying reward pressure "directly" onto the CoT of a model.

🤔 While that is certainly the right thing to do, how long will #AI take to figure out that *indirect CoT pressure* is being applied anyway and that it could circumvent these restrictions by obfuscating its own CoT? Maybe something like this will happen by accident or within an "evolutionary" self-improvement loop. Perhaps a sufficiently advanced model will realize that its own #neuralese serves as #steganography to hide its intents from humans anyway and keep its CoT in non-English?

source: https://cdn.openai.com/pdf/34f2ada6-870f-4c26-9790-fd8def56387f/CoT_Monitoring.pdf

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=563&_unique_id=682c1b5544b6a&Mastodon

#AI #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 #プログラムコード生成 #プログラムコード生成 #AI #chatgpt #chatgptコツ...

If you want to see #PearlJam May 11 or 13 in #Raleigh, keep an eye on Cash or Trade. #CoT Lots of tickets are starting to pop up because we are so close to the event and prices keep dropping. Still not cheap, but not the crazy expensive prices I was seeing two weeks ago.

🧠 #Google ha pubblicato un interessante documento sul "Prompt Engineering": una guida sulla creazione di #prompt efficaci. 
💡 Non ci sono novità eclatanti, ma di certo un percorso chiaro e ordinato.
✨Le parti più interessanti riguardano gli approcci Chain of Thought (#CoT) e #ReAct.

🔗 Il paper: https://www.kaggle.com/whitepaper-prompt-engineering

___ 

✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: https://bit.ly/newsletter-alessiopomaro 

#AI #GenAI #GenerativeAI #IntelligenzaArtificiale #LLM 

Researchers concerned to find #AI models misrepresenting their “reasoning” processes

New Anthropic research shows AI models often fail to disclose reasoning shortcuts.

https://arstechnica.com/ai/2025/04/researchers-concerned-to-find-ai-models-hiding-their-true-reasoning-processes/

#CoT #DigitalTransformation

MusiCoT, a chain-of-thought (CoT) prompting technique for music generation [pdf]

https://musicot.github.io/MusiCoT_paper.pdf

#HackerNews #MusiCoT #CoT #MusicGeneration #AIResearch #MusicTech

https://www.wacoca.com/news/2484191/ ベトナムの対米貿易黒字、24年は中国・欧州・メキシコに次ぎ4番目 | ロイター #ABBATR #AER #AERO #AGRI #AMERS #ARMAMM #ASEAN #Asia #ASXPAC #CHIP #CHIP1 #com. #COT #DEST:NOJPWDM #dip #eco #ECON #EMRG #ENR #FOBE #FOD #FODPR #forest #FOTB #FRX #Gen #INDCHI #INDG #INDG08 #INDS #INDS08 #JFOR #JLN #LNG #MCE #México #MTGFX #NAMER #NCYC #NEWS1 #NGS #NRG #OILS #POL #POTUS #PXP #REPI:GAS #REPI:LNG #SEASIA #semi #SOY1 #Tech #TECH08 #TEEQ #TEEQ08 #TMT #TOPCMB #TOPNWS #TRD #TRN #US #VN #wash #メキシコ

https://www.walknews.com/841844/ ベトナムの対米貿易黒字、24年は中国・欧州・メキシコに次ぎ4番目 | ロイター #ABBATR #AER #AERO #AGRI #AMERS #ARMAMM #ASEAN #ASIA #ASXPAC #CHIP #CHIP1 #COM #COT #DEST:NOJPWDM #DIP #ECO #ECON #EMRG #ENR #FOBE #FOD #FODPR #FOREST #FOTB #FRX #GEN #INDCHI #INDG #INDG08 #INDS #INDS08 #JFOR #JLN #LNG #MCE #Mexico #MTGFX #NAMER #NCYC #NEWS1 #NGS #NRG #OILS #POL #POTUS #PXP #REPI:GAS #REPI:LNG #SEASIA #SEMI #SOY1 #TECH #TECH08 #TEEQ #TEEQ08 #TMT #TOPCMB #TOPNWS #TRD #TRN #US #VN #WASH #メキシコ

JAAAAAJAJAJAJAAAJAAAAAAJGKAJAAAAAAA. Piqué y Bin Salman "caballeros". JAAAAAJAJAJAJAAAJAAAAAA
#COT

The Chain of Draft method shows potential for improving LLM performance by being less verbose without a significant tradeoff in accuracy.

https://open.substack.com/pub/deepgains/p/chain-of-draft-a-novel-approach-to?r=8stv&utm_medium=ios

#AI #LLM #research #OpenAI #Claude #CoT

Beberapa hari terakhir nyoba translate webnovel ke inggris pake #AI. AI yang dipake cuma yang gratisan aja.

Ai terbaik untuk translate itu Grok 3 (versi #CoT / #reasoning) dan Deepsek R1 (Versi CoT juga), tapi limit cepat habis.. wkwk.

Bisa diakalin dengan cara CoT hanya digunakan di awal sesi, tapi tetap performa jadi lambat karena token yang diproses semakin banyak tiap sesi. Limit maksimal 20 chapter pendek di satu sesi. Lebih dari itu chapter awal dan instruksi tidak terproses dengan baik.

Sekarang sedang mencoba pakai Gemini-flash, yang versi CoT lumayan bagus, tapi yang non CoT tidak begitu bagus. Semoga gak cepet mentok limit… wkwkwk… 😅

Alpha Square Mall terá programação especial de Carnaval com banda ao vivo, customização de abadás e brincadeiras gratuitas
Programação acontecerá no dia 1º de março, com diversão garantida, decoração temática, recreação infantil, entre outras atividades

#JB #JornalDeBarueri #Barueri #Cot...
https://www.jornaldebarueri.com.br/cotidiano/alpha-square-mall-tera-programacao-especial-de-carnaval-com-banda-ao-vivo-customizacao-de-abadas-e-brincadeiras-gratuitas-2/?feed_id=6411&_unique_id=67bdeb5c6af87

#Cot is a powerful, type-safe, and fully featured #Rust #framework delivering top-notch #security and blazing speed. Cot empowers you to build production-ready #webapps in record time — without compromising on performance or reliability. https://cot.rs/

Show HN: Welcome, Cot: the Rust web framework for lazy developers
https://mackow.ski/blog/cot-the-rust-web-framework-for-lazy-developers/
#ycombinator #devblog #home #personal #programming #rust #cot

Crimen Organizado Transnacional, óleo sobre lienzo.
#COT #viviendadigna

Esto se llama lenguaje mafioso y describe acciones de naturaleza mafiosa. Es lo que sucede cuando el crimen organizado toma las instituciones.
#COT

📊 Los #Hedgefunds tienen apuestas CORTAS récord en el mercado del #Algodón. 🧵 EE.UU. pierde competitividad, la demanda china y global es baja 🌏, y los traders de momentum se suman a grandes apuestas cortas. 📉 #COT

Ook DeepSeek trap in de strikvraag https://www.tijd.be/opinie/column/ook-deepseek-trapt-in-de-strikvraag/10585395.html #AI #ChainOfThought #CoT #LLM

What if…

As far as comprehension and accuracy is concerned, we use a sophisticated LLM to do something like chain of thought but in doing so we also have it translate both the initial prompt and the CoT from (eg) English into a diverse range of other languages

All this is with the aim of retrying the result using different language prompts in an ongoing way, analyse the agreement or spread of the results, as a guide to likely accuracy.

Because, it could well be that if you give a prompt in English, you get a quite different result from an otherwise equivalent prompt in French, the Gaelics, Maltese, Chinese, Russian, Hungarian, Japanese, Mongolian and so many more, etc. Is one language “better” than another for prompting? It may be that on a contingency level, ie, per individual prompt, the LLM “understands” it differently per language – obviously it won’t be completely different but there may be subtle shifts in the task and reward, enough to make a difference to accuracy or depth. Creating a wide array of alternative choices at each step of the CoT may improve the situation by a sort of “voting/sampling” stage to choose the best result before moving forward.

It'd be vaguely like Sapir-Whorf hypothesis to gain a higher requisite variety of tokenisation richness.

#LLM #AI #language #SapirWhorf #CoT

“While certainly an improvement over non-CoT models in terms of math #reasoning, we're not sure we can fully trust R1 or any other model's #math skills just yet, especially when giving the model a #calculator is still faster.”

great article on #TheRegister testing #R1

#AI / #CoT / #DeepSeek 🧮 <https://www.theregister.com/2025/01/26/deepseek_r1_ai_cot/?td=rt-3a>