AI News
Latest news and trends from the world of artificial intelligence
Google quietly released an app that lets you download and run AI models locally
Last week, Google quietly released an app that lets users run a range of openly available AI models from the AI dev platform Hugging Face on their phones. Called Google AI Edge Gallery, the app is available for Android and will soon come to iOS. It allows users to find, download, and run compatible models that generate images, answer questions, write and edit code, and more. The models run offline, without needing an internet connection, tapping into supported phones’ processors.
Micro Center nerd store fills the Fry’s vacuum with its return to Silicon Valley
Micro Center, an electronics retailer, opens its 29th store in Santa Clara, California, filling the vacuum left by Fry’s Electronics. The store offers a wide range of products for tech enthusiasts, including a Knowledge Bar, repair services, and a vast inventory of components and cutting-edge gear like AI PCs and GPUs. The store also features a section for hobbyists and DIY projects. The store's location, just minutes from Apple's headquarters, and its focus on curation and a strong inventory, prove that the once proud region still merits its own electronics store.
OpenAI sees human interaction as a competitor to ChatGPT's super assistant ambitions
An internal strategy paper reveals that OpenAI aims to develop ChatGPT into a personalized assistant by mid-2025, helping users with daily tasks, providing expertise such as programming knowledge, and performing actions online. To achieve this, OpenAI plans to use its own reasoning models and introduce new tools like "Computer Use," expanding ChatGPT's features with multimodal capabilities and generative user interfaces; a proprietary search index could serve as a key component. OpenAI wants ChatGPT to stand apart from search engines and operating systems as its own product category, identifying short-term competition from other AI chatbots and long-term competition from search services, browsers, and even human interactions.
Elevenlabs' new AI voice system enables smoother interactions through real-time analysis
Elevenlabs has released Conversational AI version 2.0, which enables AI agents to interpret conversational cues like pauses and filler sounds in real time, automatically detect spoken language, and support interactions through voice, text, or both at once. The new version integrates Retrieval Augmented Generation (RAG), allowing agents to draw directly from company databases for tasks such as medical assistance or customer support, and introduces features like automated mass calls and SIP trunking system support. For enterprise users, version 2.0 introduces compliance upgrades: Elevenlabs states the platform is fully HIPAA-compliant and offers optional EU data residency to address data protection needs.
Anthropic's Claude uses Elevenlabs technology for speech features rather than an in-house model
Programmer Simon Willison recently discovered that Anthropic lists Elevenlabs as a subcontractor for text-to-speech services in its terms of service. This suggests that, unlike OpenAI or Google, Anthropic either hasn't trained its own models on audio or hasn't achieved the necessary quality for speech synthesis on its own. Instead, Anthropic relies on technology from Elevenlabs, which just unveiled a new system aimed at making AI conversations sound more natural.
Anthropic triples revenue to $3 billion as corporate demand for generative AI surges
Anthropic now generates around $3 billion in annualized revenue, up from $1 billion in December 2024, according to Reuters sources. The sharp rise is linked to strong demand for generative AI from corporate clients. Code generation — AI tools that assist with programming — is a key growth area. Anthropic's new Claude 4 models are optimized for code generation tasks, which could drive further revenue growth. Unlike OpenAI or Google, Anthropic mainly targets business customers, suggesting that more companies are adopting AI tools.
Google says Veo 3 users have generated millions of AI videos in just a few days
Google's Veo 3 video model has been used to create millions of videos within days of its release, with the UK as the most recent launch and availability in the EU still to come.
Policy Entropy, Learning, and Alignment (Or Maybe Your LLM Needs Therapy)
I suggest therapeutic techniques from a variety of psychotherapeutic schools of thought can inspire new approaches to AI learning and alignment. I reinterpret three recent AI/ML papers in the language of psychotherapy and propose three testable training methods inspired by common psychotherapeutic interventions.
The future of engineering belongs to those who build with AI, not without it
As we look ahead, the relationship between engineers and AI systems will likely evolve from tool and user to something more symbiotic.
Idea: "Conferences as D&D tabletops"
Idea: ["Conferences as D&D tabletops"](https://gwern.net/blog/2025/conference): you may be able to better organize a conference or convention by borrowing a tool from tabletop roleplaying games - players collaborate by directly manipulating or modifying a 2D map. It seems to me like this could be low-friction and flexibly handles a lot of things that existing 'conware' design patterns don't handle well.
Archive
About Sources
AI news are automatically downloaded from various sources and translated using AI. Updates occur twice a day.