#CHainOfThought

☮ ♥ ♬ 🧑‍💻peterrenshaw@ioc.exchange
2025-06-15

“Students are introduced to advanced AI techniques such as #ChainOfThought and #SelfConsistencyPrompting, which simulate humanlike reasoning. #GenerativeAI is presented not just as a tool for queries but as a partner in reasoning.

“We teach reinforcement learning from human feedback, where every correction becomes training data,” Madmoun adds.

Students are encouraged to view AI not as a static engine, but as a responsive tool for making critical decisions in high-stakes financial environments.

Recognising that students enter with varying levels of technical knowledge, the Master in International Finance (MiF) at HEC Paris provides asynchronous #Python #programming courses, optional #BootCamps, and tailored elective tracks. “We’ve integrated workshops taught by Hi! PARIS into the curriculum,” says academic director Evren Örs, referring to the #AI and #DataScience centre co-founded by HEC Paris and Institut Polytechnique de Paris.

Students from both institutions collaborate on real-data projects, strengthening both technical and teamwork skills.

A tiered elective system requires all MiF students to complete at least one course focused on #data and #finance. The most advanced track is the #DoubleDegree in data and finance, where students dive deep into #MachineLlearning applications. Graduates, Örs says, are frequently hired as #QuantitativeAnalysts, #DataScientists, and private equity analysts in London and Paris.”

#BusinessSchools / #education <archive.md/xysyM> / <ft.com/content/071dc338-b267-4> (paywall)

Dr. Thompsonrogt_x1997
2025-06-13

🤖 Think your AI assistant can really reason? Apple’s puzzle tests say otherwise.
📉 See how “thinking” AIs collapse when logic gets real — and why we might be projecting intelligence where there is none.

Hashtags:

URL:
medium.com/@rogt.x1997/the-ill

Dr. Thompsonrogt_x1997
2025-05-31

🧠 What if your AI could explain its reasoning, verify its logic, and revise itself—all before responding?

Discover how Chain-of-Thought prompting + Self-Verification slashed hallucinations from 23% to 6%, outperformed GPT-4, and reshaped enterprise AI.

🚀 The future of trustworthy AI is here.


🔗
medium.com/@rogt.x1997/gpt-4-v

2025-05-23

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: cdn.openai.com/pdf/34f2ada6-87

If CoT pressures are used to improve agent capabilities or alignment, there may be no alternative approach to yield the same improvements without degrading monitorability. In the worst case, where the agent learns to fully obscure its intent in its CoT, we ultimately revert to the same model safety conditions that existed prior to the emergence of reasoning models and must rely on monitoring activations, monitoring potentially adversarial CoTs and outputs, or improved alignment methods. Model actions and outputs continue to grow in complexity, and we’ve demonstrated that even at their current capabilities, models can execute sophisticated reward hacks. In the near future, these models may become either too advanced or too widely used for humans to manually oversee. For this reason, we believe that CoT monitoring may become one of few tools we have for identifying latent misalignments, and it may be required that we avoid strong chain-of-thought optimization pressures entirely to guarantee we do not inadvertently cause CoT obfuscation.
Hacker Newsh4ckernews
2025-05-19
Victoria Stuart 🇨🇦 🏳️‍⚧️persagen
2025-05-02
Nick Byrd, Ph.D.ByrdNick@nerdculture.de
2025-04-21

Can popular, generalist #LLMs answer questions as specialists?

Adopting each step of #diagnosis into a #ChainOfThought prompt made small and large #languageModels' outperform both zero-shot and the fine-tuned OLAPH method on the #MedLFQA benchmark.

doi.org/10.48550/arXiv.2503.03 #AI

Structured Outputs Enable General-Purpose LLMs to be Medical Experts, pages 1 and 2.Pages 2 and 8Pages 14 and 15Pages 12 and 13
Victoria Stuart 🇨🇦 🏳️‍⚧️persagen
2025-04-10

How University Students Use Claude
anthropic.com/news/anthropic-e
news.ycombinator.com/item?id=4

Aside: been trialing SoTA LLM 😯 😀

ChatGPT, Gemini, Claude ...
counterpunch.org/2025/04/07/th

* particularly impressed w. Claude (3.7 Sonnet), DeepSeek
* most SoTA free (ChatGPT higher performing paywalled): still amazing!
* chain-of-thought reasoning / augmented responses (web retrieval: RAG) 👍️
* very impressive!!
* Firefox users: try the AI Toolbox extension 👍️

Victoria Stuart 🇨🇦 🏳️‍⚧️persagen
2025-03-31

These continue to fascinate / awe: Gemini 2.5 Pro vs. Claude 3.7 Sonnet: Coding Comparison
composio.dev/blog/gemini-2-5-p

Edit: just previewed the LLM-generated code examples - extraordinary! 🤯 😯 🥳

Victoria Stuart 🇨🇦 🏳️‍⚧️persagen
2025-03-29
Victoria Stuart 🇨🇦 🏳️‍⚧️persagen
2025-03-29

/1 That post includes the following video - which in simple language / examples a basic overview of a current, reasoning LLM (thought; Claude) and "prompt engineering."

Tracing the thoughts of a large language model
youtube.com/watch?v=Bj9BD2D3DzA

Winbuzzerwinbuzzer
2025-03-09

Zoom has introduced Chain of Draft, a new AI prompting method that reduces token usage by 92% and slashes operational costs by 90%

winbuzzer.com/2025/03/09/zooms

Client Info

Server: https://mastodon.social
Version: 2025.04
Repository: https://github.com/cyevgeniy/lmst