stp2y - CybAI news

23 May

AI Is a Black Box. Anthropic Figured Out a Way to Look Inside

stp2y0 CommentsAIalgorithms, artificial intelligence, deep learning, neural networks, research

Last year, the team began experimenting with a tiny model that uses only a single layer of neurons. (Sophisticated LLMs have dozens of layers.) The hope was that in the simplest possible setting they could discover patterns that designate features. They ran countless experiments with no success. “We tried a whole bunch of stuff, and nothing was working. It looked like a bunch of random garbage,” says Tom Henighan, a member of Anthropic’s technical staff. Then a run dubbed “Johnny”—each experiment was assigned a random name—began associating neural patterns with concepts that appeared in its outputs.“Chris looked at it, and…

23 May

ChatGPT is now better than ever at faking human emotion and behaviour

stp2y0 CommentsChat-GPT

Earlier this week OpenAI launched GPT-4o (“o” for “omni”), a new version of the artificial intelligence (AI) system powering the popular ChatGPT chatbot. GPT-4o is promoted as a step towards more natural engagement with AI. According to the demonstration video, it can have voice conversations with users in near real-time, exhibiting human-like personality and behaviour. This emphasis on personality is likely to be a point of contention. In OpenAI’s demos, GPT-4o sounds friendly, empathetic and engaging. It tells “spontaneous” jokes, giggles, flirts and even sings. The AI system also shows it can respond to users’ body language and emotional tone.…

23 May

AI-powered metadata: Informatica’s role in the future of data management – SiliconANGLE

stp2y0 CommentsGenAIAI-powered metadata: Informatica’s role in the future of data management, Guest Author, SiliconANGLE

The journey to AI-integrated metadata systems As theCUBE Research has been tracking, the transformation of data management systems has been gradual yet profound. Over the past decade, organizations have increasingly recognized the value of separating computational power from data storage, leading to more centralized data consolidation and management. Informatica Inc., a data integration and management leader since the 1990s, has evolved into a comprehensive data management platform that integrates artificial intelligence and metadata to enhance data utilization across various industries. Challenges of using AI in metadata integration As we have discussed in past articles about the Sixth Data Platform and…

23 May

Predictive Human Preference: From Model Ranking to Model Routing

stp2y0 CommentsRAG models

A challenge of building AI applications is choosing which model to use. What if we don’t have to? What if we can predict the best model for any prompt? Predictive human preference aims to predict which model users might prefer for a specific query. Table of contents Ranking Models Using Human Preference…. How Preferential Ranking Works…. Correctness of Chatbot Arena Ranking…….. Eval data…….. ResultsPredicting Human Preference For Each Prompt…. Experiment setup…. Experiment results…….. Domain-specific and query-specific leaderboardsConclusion Human preference has emerged to be both the Northstar and a powerful tool for AI model development. Human preference guides post-training techniques including…

23 May

GenAI Just Got A Little Less Opaque

stp2y0 CommentsNews

Yesterday, the AI startup Anthropic published a paper detailing the successful interpretation of the inner workings of a large language model (LLM). LLMs are notoriously opaque — their size, complexity, and numeric representation of human language have hitherto defied explanation — so it’s impossible to understand why inputs lead to outputs. Anthropic used a technique called dictionary learning, leveraging a sparse encoder to isolate specific concepts within its Claude 3 Sonnet model. The technique allowed them to extract millions of features, including specific entities like the Golden Gate Bridge as well as more abstract ideas such as gender bias. They…

23 May

Generating fashion product descriptions by fine-tuning a vision-language model with SageMaker and Amazon Bedrock | Amazon Web Services

stp2y0 CommentsAI

In the world of online retail, creating high-quality product descriptions for millions of products is a crucial, but time-consuming task. Using machine learning (ML) and natural language processing (NLP) to automate product description generation has the potential to save manual effort and transform the way ecommerce platforms operate. One of the main advantages of high-quality product descriptions is the improvement in searchability. Customers can more easily locate products that have correct descriptions, because it allows the search engine to identify products that match not just the general category but also the specific attributes mentioned in the product description. For example,…

23 May

Tech policy is only frustrating 90% of the time

stp2y0 CommentsChat-GPT

Many technologists stay far away from public policy. That’s understandable. In our experience, most of the time when we engage with policymakers there is no discernible impact. But when we do make a difference to public policy, the impact is much bigger than what we can accomplish through academic work. So we find it fruitful to engage even if it feels frustrating on a day-to-day basis.In this post, we summarize some common reasons why many people are cynical about tech policy and explain why we’re cautiously optimistic. We also announce some recent writings on tech policy as well as an…

23 May

Inside the Pod: A Guide to Launching a Successful One-person Company

stp2y0 CommentsRAG models

Thank you to everyone who is watching or listening to my podcast, How Do You Use ChatGPT? If you want to see a collection of all of the prompts and responses in one place, Every contributor Rhea Purohit is breaking them down for you to replicate. Let us know in the comments if you find these guides useful. —Dan ShipperWas this newsletter forwarded to you? Sign up to get it in your inbox.In the first week of April, I came face-to-face with the part of being a freelancer I hate the most: tax season.The idea of wading through invoices, account…

23 May

Scale AI

stp2y0 CommentsNews

I’m excited to share Generational’s inaugural growth & late-stage company briefing with a deep dive on Scale AI, blending analytical rigor with feature writing. Disclaimer: I have a financial interest in Scale. Don’t take this as investment advice.In this deep dive, you’ll learn insights from conversations with Scale’s customers, ex-employees, and competitors. I could do this thanks to Tegus, which centralizes expert calls into a single platform. Nothing beats primary research when it comes to understanding a company. If you’re curious about Tegus, try them out with this link.Scale AI accelerates the development of AI applications through services and software.…

23 May

Poking parts of Sonnet’s brain to make it less annoying

stp2y0 CommentsAI

In a groundbreaking new paper (actually groundbreaking, IMO), researchers at Anthropic have scaled up an interpretability technique called "dictionary learning" to one of their deployed models, Claude 3 Sonnet. The results provide an unprecedented look inside the mind of a large language model, revealing millions of interpretable features that correspond to specific concepts and behaviors (like sycophancy) and shedding light on the model's inner workings. In this post, we'll explore the key findings of this research, including the discovery of interpretable features, the role of scaling laws, the abstractness and versatility of these features, and their implications for model steering…