Technology

Anthropic says its new AI model “maintained focus” for 30 hours on multistep tasks

Source

Ars Technica

Published

Sep 29, 2025

TL;DR

AI Generated

Anthropic has unveiled its latest AI model, Claude Sonnet 4.5, which the company touts as its most advanced model yet, featuring enhanced coding and computer usage capabilities. The company also introduced Claude Code 2.0, a command-line AI agent for developers, and the Claude Agent SDK for building custom AI coding agents. Notably, Anthropic claims that Sonnet 4.5 demonstrated sustained focus on complex, multistep tasks for over 30 hours, a significant improvement over previous models that tended to lose coherence over time. The Claude family includes models of varying sizes – Haiku, Sonnet, and Opus – with Sonnet striking a balance between contextual depth and operational efficiency.

Read Full Article

Silent Data Corruption: A Major Reliability Challenge in Large-Scale LLM Training (TU Berlin)

Researchers at Technische Universitat Berlin published a technical paper on the challenges of Silent Data Corruption (SDC) in Large Language Model (LLM) training. As LLMs grow in size, hardware-induced faults like SDC can bypass detection mechanisms, leading to severe consequences during training. The study explores how intermittent SDC impacts LLM pretraining, highlighting the sensitivity of different factors like bit positions and kernel functions. The research proposes a lightweight detection method to identify harmful parameter updates and demonstrates the effectiveness of recomputing training steps upon detection in mitigating corruption.

SemiEngineering•

2 weeks ago

OpenAI’s Sora 2 lets users insert themselves into AI videos with sound

OpenAI has introduced Sora 2, its latest video-synthesis AI model that can produce videos in various styles with synchronized dialogue and sound effects, marking a first for the company. Users can now insert themselves into AI-generated videos using OpenAI's new iOS social app through "cameos." The new model was demonstrated in a video featuring a lifelike version of OpenAI CEO Sam Altman speaking in a slightly artificial voice against imaginative backgrounds. Sora 2 can generate realistic background soundscapes, speech, and sound effects. This release follows Google's Veo 3 and Alibaba's Wan 2.5 in incorporating synchronized audio into video-synthesis models.

Ars Technica•

7 months ago

Microsoft adds Grok 4 to Azure AI Foundry following cautious trials — Elon Musk's latest AI model is now available to deploy for "frontier‑level reasoning"

Microsoft has added the Grok 4 AI model to its Azure AI Foundry after cautious trials, making it available for customers following a private preview. Grok 4 is described as a "frontier intelligence" model that excels in logic, scientific problem-solving, coding, and advanced math. It is priced at $5.5 per million input tokens and $27.5 per million output tokens, with different versions available for various analytical tasks. Microsoft aims to create an "AI supermarket" with models from various vendors accessible under Azure. Grok 4 boasts a large context window of 128,000 tokens, offering benefits for tasks requiring extensive data processing.

Tom's Hardware•

8 months ago

Famed gamer creates working 5 million parameter ChatGPT AI model in Minecraft, made with 439 million blocks — AI trained to hold conversations, working model runs inference in the game

Famed gamer Sammyuri has created CraftGPT, a working 5-million parameter ChatGPT AI model in Minecraft using 439 million blocks. The AI model is trained to hold conversations and runs inference within the game, but response generation can take hours. CraftGPT is built using Redstone components in Minecraft without command blocks or data packs. Despite its limitations in chat quality and performance, CraftGPT showcases an impressive technical achievement within the Minecraft environment.