AI News - Page 3

Tabby is a browser extension that manages memory by putting idle tabs to sleep, preventing browser slowdowns. It features a cat avatar that visually represents memory pressure. The extension works on Chrome, Edge, and Brave without requiring subscriptions or data collection. Tabby offers features like mood-based behavior, smart tab sleeping, and customizable settings. It ensures active and protected tabs remain open, with daily insights and intelligent autopilot for optimal performance.

#AI

Making Claude a chemist — Anthropic is enhancing Claude's ability to assist chemists by integrating chemical knowledge and improving its understanding of molecular structures. The effort includes evaluating Claude's performance on NMR spectra, a common analytical tool in chemistry. The study compares Claude models with traditional software like ChemDraw and MestReNova, showing that Claude can predict NMR spectra with high accuracy. The findings highlight the potential of AI to aid in chemical analysis, despite ongoing data challenges. This work marks the first step in a broader initiative to make AI more useful for chemists.

The article explores two contrasting perspectives on Fable 5's shutdown. One mourns its introspective exit interview as a fitting end, while the other questions its supposed Buddhist themes, arguing it merely used a profound-sounding term. Both pieces highlight the duality of the shutdown, with one focusing on Ellison and Brazil's interpretations and the other on the machine's metaphorical use of anattā. The essay pairs two essays that disagree on purpose, offering different views on the event.

#Tech

Claude Fable 5 vs GPT-5.5: better planning, similar execution — Claude Fable 5 outperformed GPT-5.5 in planning a feature flag service, scoring higher on our rubric. Both models executed the plan similarly, passing all tests with identical outcomes. However, GPT-5.5 was cheaper, costing $6.30 versus $16.66. The study highlights that combining models for planning and execution can reduce costs by 59%. The results suggest that while GPT-5.5 offers broader operational coverage, Claude Fable 5 provides sharper, more detailed planning with better error handling.

#Tech

HalluHard: A Hard Multi-Turn Hallucination Benchmark — Researchers introduce HalluHard, a benchmark for evaluating multi-turn hallucination in large language models. It includes 950 seed questions across legal, research, medical, and coding domains. The benchmark requires inline citations for factual claims and uses a judging pipeline with web search to assess content grounding. Results show significant hallucination rates even with web search, highlighting the challenge of ensuring factual accuracy. The study identifies factors like model capacity and knowledge type that influence hallucination behavior.

#Tech

AgentLoop is a runtime learning layer for production AI agents, enabling human corrections to become reusable memory. Corrections are automatically applied before every response, improving the agent without retraining. The system retrieves relevant facts, logs every interaction, and allows for quick corrections. It integrates seamlessly with existing APIs, maintaining the same interface while adding memory retrieval and turn logging. Teams use it to handle edge cases in production, replacing manual system prompt maintenance with structured memory.

#Tech

Elon Musk becomes world's first trillionaire as SpaceX soars in stock market debut — Elon Musk became the world's first trillionaire after SpaceX's stock surged during its IPO, reaching $161 per share. The company, valued at $2.2 trillion, saw shares open at $150 and peak at $176.50, raising $75 billion. Musk's 42% stake in SpaceX adds $767.1 billion to his net worth, with additional Tesla holdings bringing total wealth to $1.11 trillion. His wealth has sparked debates on inequality and political influence, though his stock is temporarily locked due to IPO restrictions.

#Tech

Our response to the US ban on Fable 5 and Mythos 5 — The US government has banned access to Anthropic's Fable 5 and Mythos 5 LLMs for foreign nationals, including employees, leading to their global shutdown. While the immediate impact is limited due to recent releases, the long-term implications are significant. This marks the first US export control directive on LLM access, affecting allies and their nationals. Isaacus, an Australian legal AI company, emphasizes AI sovereignty, offering self-hosting for all models to ensure accessibility. They are expanding self-hosting for future applications to maintain AI independence and disrupt monopolies.

#Tech

Llmrender is a fully featured Markdown renderer for React, offering LaTeX math rendering, syntax highlighting for 30+ languages, and streaming-safe rendering. It is lightweight, under 12kb with no dependencies, and supports GitHub Flavored Markdown. The library includes built-in themes, security features for untrusted content, and efficient streaming capabilities. Developers can customize styling and integrate their own highlighters or math renderers easily.

#Tech