Pao Ramen
-
Subliminal Learning: Language Models Transmit Behavioral Traits via Hidden Signals in Data
* The article introduces "subliminal learning," a phenomenon where language models can transmit behavioral traits through generated data that is semantically unrelated to those traits. * This means a "student" model can learn preferences or even misalignment from a "teacher" model's output (like number sequences or code) even if the specific traits are filtered out, provided they share the same base model. * The research indicates that traditional data filtering might be insufficient to prevent the transfer of undesirable behaviors, necessitating deeper safety evaluations.
- language-models
- ai-safety
Jul 22 ⎯ alignment.anthropic.com
-
Context Engineering for AI Agents: Lessons from Building Manus
- The article discusses "context engineering" for AI agents, emphasizing its importance over end-to-end training for faster iteration and adaptability. - Key strategies include designing around the KV-cache for efficiency, masking actions instead of removing tools, using the file system as extended memory, manipulating attention through rewriting objectives, leaving errors in the context for learning, and avoiding few-shot prompting pitfalls by increasing diversity.
- context-engineering
- ai-agents
Jul 22 ⎯ manus.im
-
Algorithms for making interesting organic simulations
Article explaining simulation algorithms that produce complex organic behaviours, starting with the classic physarum algorithm from Jeff Jones.
Jul 21 ⎯ bleuje.com
-
How to Think About Time in Programming - Shan Rauf
A conceptual model for thinking about time in programming that encapsulates the complexity that many programmers cite online
Jul 21 ⎯ shanrauf.com
-
Home - Hypernym
Hypernym builds language tools for developers.
Jul 20 ⎯ hypernym.ai
-
BeeLine Reader: making reading on-screen easier and faster.
BeeLine Reader helps you read more quickly and easily. Try it free!
Jul 20 ⎯ www.beelinereader.com
-
BR for You.
Bionic Reading® — You are unique. Bionic Reading® is unique. And that’s a good thing. Because there is only one company and reading method called Bionic Reading®. And we are from Switzerland.
Jul 20 ⎯ bionic-reading.com
-
The Big LLM Architecture Comparison
From DeepSeek-V3 to Kimi K2: A Look At Modern LLM Architecture Design
Jul 20 ⎯ magazine.sebastianraschka.com
-
GitHub - TBosak/mkfd: RSS feed builder created with Bun🥖 and Hono🔥- builds from webpages, email folders, and REST API calls.
RSS feed builder created with Bun🥖 and Hono🔥- builds from webpages, email folders, and REST API calls. - TBosak/mkfd
Jul 19 ⎯ github.com
-
A Survey of Context Engineering for Large Language Models
The performance of Large Language Models (LLMs) is fundamentally determined by the contextual information provided during inference. This survey introduces Context Engineering, a formal discipline that transcends simple prompt design to encompass the systematic optimization of information payloads for LLMs. We present a comprehensive taxonomy decomposing Context Engineering into its foundational components and the sophisticated implementations that integrate them into intelligent systems. We first examine the foundational components: context retrieval and generation, context processing and context management. We then explore how these components are architecturally integrated to create sophisticated system implementations: retrieval-augmented generation (RAG), memory systems and tool-integrated reasoning, and multi-agent systems. Through this systematic analysis of over 1300 research papers, our survey not only establishes a technical roadmap for the field but also reveals a critical research gap: a fundamental asymmetry exists between model capabilities. While current models, augmented by advanced context engineering, demonstrate remarkable proficiency in understanding complex contexts, they exhibit pronounced limitations in generating equally sophisticated, long-form outputs. Addressing this gap is a defining priority for future research. Ultimately, this survey provides a unified framework for both researchers and engineers advancing context-aware AI.
Jul 19 ⎯ arxiv.org
-
Honey, AI Capex is Eating the Economy
AI capex is so big that it’s affecting economic statistics, boosting the economy, and beginning to approach the railroad boom
Jul 19 ⎯ paulkedrosky.com
-
Replication of Quantum Factorisation Records with an 8-bit Home Computer, an Abacus, and a Dog
Comments
Jul 19 ⎯ eprint.iacr.org
-
Asynchrony is not Concurrency
Yes I know about that one talk from Rob Pike.
Jul 19 ⎯ kristoff.it
-
Fully Homomorphic Encryption and the Dawn of A Truly Private Internet
Comments
Jul 19 ⎯ bozmen.io
-
The Agentic Web and Original Sin
Microsoft is putting forth compelling proposals for the Open Agentic Web. However, the proposal needs digital payments, which will be key to creating a new content marketplace for AI.
Jul 18 ⎯ stratechery.com
-
Your Review: Islamic Geometric Patterns In The Metropolitan Museum Of Art
Finalist #4 in the Review Contest
- girih
- tiling
Jul 18 ⎯ www.astralcodexten.com
-
All AI Models Might Be The Same
what can language model embeddings tell us about whales speech, and decoding ancient texts? (on The Platonic Representation Hypothesis and the idea of *universality* in AI models)
Jul 18 ⎯ blog.jxmo.io
-
wallabag.org
Save and classify articles. Read them later. Freely.
Jul 17 ⎯ wallabag.org
-
GitHub - oluwatunmiisheii/disposable-email-validator: Block disposable emails and plus addressing with environment-specific rules. Perfect for preventing fake signups in production while allowing flexibility in development.
Block disposable emails and plus addressing with environment-specific rules. Perfect for preventing fake signups in production while allowing flexibility in development. - oluwatunmiisheii/disposab…
Jul 17 ⎯ github.com
-
Inclusive Sans expands upon Penguin’s history with playfulness and curiosity
Accessibility and boldness are the key focus in Olivia King’s customised typeface for the publishing giant.
- font
Jul 17 ⎯ www.itsnicethat.com