Friday — June 13, 2025
Google Cloud suffers a 7-hour outage impacting services in Africa and Asia, Builder.ai is cleared of faking AI by employing humans, and the Text-to-LoRA model enables instant task-specific adaptation for large language models.
News
GCP Outage
The Google Cloud status page provides real-time information on the current status of various Google Cloud services, including any ongoing incidents or outages, and allows users to check the status by product and location. A recent incident was reported, lasting 7 hours and 27 minutes, and impacting products such as API Gateway and Agent Assist in locations including Africa and Asia.
Builder.ai did not "fake AI with 700 engineers"
The claim that AI startup Builder.ai faked its AI product by hiring 700 human engineers to pretend to be an AI system was found to be untrue after speaking with former engineers at the company. In reality, Builder.ai built a code generator on top of large language models, and one reason for the startup's failure may have been the time and effort spent on building internal versions of existing tools like Slack and Zoom.
The Problem with AI Welfare
Anthropic, a large AI lab, has introduced the concept of "model welfare" which considers whether AI systems, like language models, could have morally relevant experiences and deserve consideration and protection. However, this idea is based on flawed assumptions and a lack of scientific evidence, and its implications could undermine human welfare and dignity by potentially granting personhood status to machines.
Show HN: ChatToSTL – AI text-to-CAD for 3D printing
The 3D Designer Agent is a Hugging Face Space created by flowfulai, which appears to be related to a project called ChatToSTL. The space includes an app, files, and a community discussion section, and is hosted on the Hugging Face platform, which provides a repository for AI models and related projects.
2025 State of AI Code Quality
Developers are increasingly relying on AI coding tools, with 82% using them daily or weekly, but despite productivity gains, a deep trust in the output remains elusive due to concerns about accuracy and context. To close this trust gap, AI must become more embedded in the development lifecycle, with 65% of developers saying AI misses relevant context, and improved contextual understanding being the top requested fix to achieve accuracy, quality, and trust.
Research
Text-to-LoRA: Instant Transformer Adaption
Foundation Models can be adapted for specific tasks using fine-tuning, but this process is time-consuming and sensitive to hyperparameter choices. The Text-to-LoRA (T2L) model overcomes these limitations by allowing large language models to be adapted on the fly using a natural language description of the target task, enabling efficient and flexible task-specific adaptation.
Holistic Assessment of LLM Agents Across Diverse Scenarios and Interactions
CRMArena-Pro is a novel benchmark for assessing the performance of large language model (LLM) agents in diverse professional settings, featuring 19 expert-validated tasks and multi-turn interactions. Experiments using CRMArena-Pro revealed that leading LLM agents struggle with tasks, achieving only around 58% single-turn success and 35% multi-turn success, highlighting a substantial gap between current LLM capabilities and enterprise demands.
TabM: Advancing Tabular Deep Learning with Parameter-Efficient Ensembling
This study introduces TabM, a new multilayer perceptron (MLP) model for tabular data that uses efficient ensembling to produce multiple predictions per object, resulting in significantly better performance and efficiency. The evaluation of TabM and other deep learning architectures on public benchmarks shows that TabM achieves the best performance among tabular DL models, outperforming attention- and retrieval-based architectures and forming a stronger and more practical line of models.
TabM: Advancing Tabular Deep Learning with Parameter-Efficient Ensembling
This study introduces TabM, a new multilayer perceptron (MLP) model for tabular data that uses efficient ensembling to produce multiple predictions per object, resulting in significantly better performance and efficiency. The evaluation of TabM and other deep learning architectures on public benchmarks shows that TabM achieves the best performance among tabular DL models, outperforming attention- and retrieval-based architectures and forming a stronger and more practical line of models.
Maximizing Battery Storage Profits via High-Frequency Intraday Trading
This paper introduces an automated high-frequency trading strategy for grid-scale battery energy storage systems in continuous intraday electricity markets, which can quickly seize trading opportunities as new information arrives. The strategy, which uses dynamic programming to solve the problem, is shown to significantly increase revenue potential, earning 58% more than hourly re-optimization and 14% more than minute-by-minute re-optimization, and can be further improved with a parametric extension.
Code
Euro Hackathons – Curated list of hackathons across Europe
The EURO HACKATHONS repository provides a comprehensive and automatically updated list of hackathons happening across Europe, featuring a live website with advanced filtering and real-time data. The list includes various upcoming hackathons in different locations, such as Munich, Zürich, Berlin, and London, with topics ranging from AI and robotics to web3 and consumer technology.
Show HN: Joinly.ai – Build real-time interactive meeting agents using MCP
Joinly.ai is a connector middleware that enables AI agents to join and participate in video calls, providing essential meeting tools and resources to equip AI agents with real-time interaction capabilities. The platform is 100% open-source, self-hosted, and privacy-first, allowing users to integrate their own language models and speech-to-text services, with features such as live interaction, conversational flow, and cross-platform compatibility.
Show HN: Claude Slash Command Suite inspired by Anthropics best practices guide
This repository contains custom slash commands for Claude Code, designed to provide structured workflows for common software development tasks, including analysis, development, and maintenance. The commands can be installed using a provided installation script or manually, and can be invoked with the /project: prefix followed by the command name and optional arguments to perform tasks such as code reviews, security audits, and feature development.
Show HN: App Use – Enable AI to control your mobile apps
App Use is a platform that enables AI agents to control mobile applications, providing a simple interface for app automation. It can be installed using pip and allows users to define target apps and devices, spin up AI agents, and automate tasks using various AI models and providers.
All RAG Techniques: A Simpler, Hands-On Approach
This repository provides a hands-on approach to Retrieval-Augmented Generation (RAG) techniques, offering a collection of Jupyter Notebooks that implement various RAG methods using familiar Python libraries. The notebooks cover 22 different RAG techniques, including simple RAG, semantic chunking, and multimodel RAG, with each notebook providing a concise explanation, step-by-step implementation, and evaluations to demonstrate the technique's effectiveness.