GPT 4.5 level for 1% of the price

Baidu Inc. has unveiled ERNIE 4.5 and X1, with ERNIE X1 offering performance comparable to DeepSeek R1 at half the price, and ERNIE 4.5 being the latest foundation model with multimodal capabilities. The company has also made its AI chatbot ERNIE Bot free to individual users, with both models accessible via the official website.

Read more

AI Is Making Developers Dumb

The author argues that relying on Large Language Models (LLMs) like GitHub Copilot can make developers "dumber" by reducing their ability to think critically and understand the underlying code, leading to a loss of foundational knowledge and self-sufficiency. However, the author also suggests that LLMs can be useful tools for research and learning when used with a critical and inquisitive mindset, encouraging developers to interrogate and understand the recommendations rather than simply trusting the output.

Read more

"Wait, not like that": Free and open access in the age of generative AI

The open access movement, which aims to make knowledge and education freely available, is facing challenges in the age of generative AI, as some creators are having "wait, no, not like that" moments when they see their work being used in ways they didn't intend. In response, some are trying to regain control by tightening licenses or limiting access, but this risks undermining the very purpose of free and open access, and may not even be effective in preventing unwanted uses.

Read more

Leaked Apple meeting shows how dire the Siri situation is

Apple's Siri team is facing significant challenges, with delayed AI-powered features announced last June potentially not making it into iOS 19, and the company's senior director Robby Walker calling the situation "ugly" and "embarrassing". The delays have been attributed to quality issues, with the features not working properly up to a third of the time, and the company is now aiming to ship them as soon as they are ready, with no guaranteed timeline.

Read more

Deepseek reportedly restricts employee travel amid AI security concerns

Chinese AI company Deepseek has reportedly restricted employee travel amid AI security concerns, with some employees required to surrender their passports. The move is likely aimed at preventing data leaks and unauthorized acquisitions, and comes as the Chinese government increases its scrutiny of the AI sector, advising top AI entrepreneurs and researchers to limit their travel to the US and other countries.

Read more

Sketch-of-Thought: Efficient LLM Reasoning

The Sketch-of-Thought (SoT) framework is a novel prompting approach that combines cognitive-inspired reasoning with linguistic constraints to reduce token usage in large language models while maintaining reasoning accuracy. SoT achieves a 76% reduction in tokens with minimal impact on accuracy, and in some cases, it even improves accuracy while using fewer tokens, as demonstrated through evaluations across 15 reasoning datasets.

ArXiv

Operationalizing Machine Learning: An Interview Study

Organizations rely on machine learning engineers to deploy and maintain ML pipelines in production through a process called MLOps, which involves a continuous loop of data collection, experimentation, evaluation, and monitoring. Researchers conducted interviews with 18 machine learning engineers to identify key challenges and best practices, revealing three crucial variables for success: Velocity, Validation, and Versioning, and highlighting areas for improvement in tool design.

ArXiv

AI and the value of privacy-preserving tools to distinguish who is real online

The increasing capabilities of AI have made it easier for malicious actors to conduct deceptive schemes online, making it challenging to balance anonymity and trustworthiness. Personhood credentials, digital credentials that verify a user is a real person without disclosing personal information, offer a potential solution to address this challenge and reduce misuse by bad actors on online platforms.

ArXiv

Machine Learning Operations (MLOps): Overview, Definition, and Architecture

The goal of industrial machine learning projects is to develop and deploy ML products quickly, but many projects fail due to difficulties in automation and operationalization, which is addressed by the Machine Learning Operations (MLOps) paradigm. Through research, including literature reviews and expert interviews, a comprehensive overview of MLOps principles, components, and workflows is provided, along with a definition and guidance for ML researchers and practitioners to successfully automate and operate their ML products.

ArXiv

Do Emotions Affect Argument Convincingness?

Emotions can influence the convincingness of arguments, but their impact is not always significant, with over half of cases showing no change in convincingness despite variations in emotional intensity. When emotions do have an effect, they tend to enhance convincingness rather than weaken it, a pattern that is generally mirrored by large language models (LLMs), although they struggle to capture nuanced emotional effects in individual judgments.

ArXiv

Docs – Open source alternative to Notion or Outline

Docs is an open-source document editor that enables live collaboration and allows users to turn their notes into knowledge through simple and secure editing features. The platform offers various tools, including offline editing, AI-powered actions, and granular access control, making it a scalable and secure alternative to other collaboration tools like Notion, Outline, or Confluence.

Repo

Show HN: Computer – Build Your Manus AI Agent with an OSS macOS Sandbox

Cua is a project that allows users to create and run high-performance macOS and Linux VMs on Apple Silicon, with built-in support for AI agents, and provides various libraries and tools for interacting with these VMs. The project includes libraries such as Lume, Computer, and Agent, which can be installed using brew or pip, and offers documentation, demos, and a community Discord channel for support and discussion.

Repo

Show HN: Create a local RAG AI in 2 minutes

RLAMA is a powerful AI-driven question-answering tool that enables users to create, manage, and interact with Retrieval-Augmented Generation (RAG) systems tailored to their documentation needs. The tool seamlessly integrates with local Ollama models and supports various features, including document processing, vector storage, and context retrieval, with a roadmap that includes future enhancements such as advanced embedding pipelines, user experience improvements, and enterprise features.

Repo

Python Playwright E2E tests with the right amount of AI (almost none)

Playsmart is a Python library that uses Playwright and OpenAI to automate end-to-end testing by allowing users to write tests in a more human-like language, such as "click on login" or "fill email input with hello@world.tld". The library uses a caching layer to reduce the number of requests made to the OpenAI API and can be installed via PyPI with the command pip install playsmart.

Repo

VimKiller Recharged plus AI

VIMKiller RECHARGED is an open-source software project that uses AI to help users exit the VI(M) text editor, which can be frustrating to exit. The project uses a voice-activated system, where users can say a customizable code word, such as "The octopus has escaped", to exit VI(M), and is available for donation, with suggested donations ranging from $250,000 to $777,777.

Repo