Apple restricts Pebble from being awesome with iPhones

The developer of a smartwatch is warning that their iOS app will have limited functionality compared to the Android version due to Apple's restrictions on third-party smartwatches, which prevent features like sending text messages, replying to notifications, and integrating with other iOS apps. The situation has allegedly worsened over the years, with Apple imposing more restrictions, and the developer is urging iPhone users to complain or switch to Android to pressure Apple to change its policies.

Read more

FTC Removes Posts Critical of Amazon, Microsoft, and AI Companies

The Federal Trade Commission (FTC) has removed over 300 blogs from its website, including posts critical of Amazon, Microsoft, and AI companies, as well as consumer protection information related to artificial intelligence and privacy lawsuits. The removed content was published during the Biden administration, and its removal has raised concerns about the loss of important guidance and information on consumer protection and technology issues.

Read more

Two new PebbleOS watches

Two new smartwatches, the Core 2 Duo and Core Time 2, are being released, both of which run on open source PebbleOS and are compatible with thousands of existing Pebble apps. The Core 2 Duo features a black and white display, polycarbonate frame, and costs $149, while the Core Time 2 has a larger 64-color display, metal frame, and costs $225, with shipping starting in July and December, respectively.

Read more

AI crawlers haven't learned to play nice with websites

SourceHut, a git-hosting service, is experiencing disruptions due to aggressive web crawlers from AI companies, which are slowing down its services with excessive demands for data. The company has deployed mitigations, including blocking certain cloud providers, to contain the issue, but notes that this may impact end-users, highlighting a growing problem of ill-behaved AI crawlers overwhelming websites with requests.

Read more

Hungary's use of facial recognition violates EU AI Act

Hungary's proposed use of facial recognition technology to monitor and identify participants in pride events violates the EU's AI Act, which prohibits the use of live facial recognition in public spaces except in exceptional circumstances. The Hungarian government's plan to ban pride events and use facial recognition to enforce the ban has been criticized by MEPs and NGOs, who argue that it infringes on the right to assembly and personal data protection.

Read more

Politicians' misinformation behavior and public engagement, in 4 countries

A study of 3,277 politicians across Germany, Italy, the UK, and the USA found significant cross-country differences in misinformation-sharing behavior, with politicians in Italy and the USA exhibiting the highest rates, particularly among far-right and conservative legislators. The study also revealed varying levels of public engagement with misinformation, with the USA showing significantly higher engagement with misinformation than reliable information, and Italy having uniquely high levels of crisis-related misinformation, particularly regarding COVID-19.

ArXiv

Reinforcement Learning via Self-Play

Recent AI advancements are transforming large language models (LLMs) into large reasoning models (LRMs) that can perform complex reasoning tasks, and researchers have proposed a post-training framework called Reinforcement Learning via Self-Play (RLSP) to enable search in LLMs. The RLSP framework has shown promising results, improving reasoning performance in math domains and exhibiting emergent behaviors such as backtracking and exploration, suggesting its potential to enable complex reasoning abilities in LLMs when scaled.

ArXiv

EXAONE Deep: Reasoning Enhanced Language Models

The EXAONE Deep series of models has demonstrated superior performance in various reasoning tasks, including math and coding benchmarks, after being trained on a specialized dataset. The models, which range in size from 2.4B to 32B, have outperformed comparable models and are openly available for research purposes.

ArXiv

Blaze: Compiling JSON Schema for 10x Faster Validation

JSON Schemas are crucial for validating the structure of JSON input in Web APIs, but the validation process can be time-consuming and add overhead to requests. The Blaze JSON Schema validator addresses this issue by compiling complex schemas into an efficient representation, reducing validation time by an average of 10x compared to existing validators while maintaining strict adherence to the JSON Schema specification.

ArXiv

Chain of Draft: Thinking Faster by Writing Less

Chain of Draft (CoD) is a new approach that enables Large Language Models (LLMs) to generate concise, informative intermediate reasoning outputs, mimicking human cognitive processes. CoD achieves comparable or better accuracy than Chain-of-Thought (CoT) prompting while using significantly fewer tokens, reducing cost and latency in various reasoning tasks.

ArXiv

Nvidia Dynamo: A Datacenter Scale Distributed Inference Serving Framework

NVIDIA Dynamo is a high-throughput, low-latency inference framework designed for serving generative AI and reasoning models in distributed environments, offering features such as disaggregated prefill and decode inference, dynamic GPU scheduling, and accelerated data transfer. The framework is open-source, built in Rust and Python, and provides a simple way to spin up local inference components, including an OpenAI compatible frontend, router, and workers, to serve large language models (LLMs) like those from Hugging Face.

Repo

Show HN: Zant – A TinyML SDK in Zig

Zant (Zig-Ant) is an open-source SDK for deploying optimized neural networks on microcontrollers, addressing the lack of comprehensive deep learning libraries on devices like TI Sitara and Raspberry Pi Pico. The project offers real-time optimizations, cross-platform compatibility, and modular integration, with a roadmap that includes short-term goals like MNIST inference and YOLO implementation, as well as long-term goals like advanced pruning and expanded microcontroller compatibility.

Repo

Anubis: Weighs the soul of HTTP requests using proof-of-work to stop AI crawlers

Anubis is a proof-of-work challenge system that protects upstream resources from scraper bots by requiring clients to calculate a SHA-256 checksum, with a customizable difficulty level, before accessing a website. The system sets an HTTP cookie with a signed JSON Web Token (JWT) when a client passes the challenge, which contains metadata to prove the token's validity and can be used to bypass future challenges.

Repo

Block's Open Source AI agent Goose's system prompt

Goose is an open-source AI agent that automates complex engineering tasks, from building projects to debugging and interacting with external APIs, allowing developers to focus on innovation. It is designed to be flexible, working with any large language model and integrating with various APIs, and offers extensive documentation and community support through multiple channels.

Repo

LangGraph Builder

LangGraph Builder is a tool that provides a canvas for designing cognitive architectures of LangGraph applications and generates boilerplate code in Python and Typescript. It can be tried out online or run locally by installing dependencies and starting a development server, although it has some limitations, including a lack of mobile support and no support for parallel node execution.

Repo