AI News

Follow the latest AI and automation news

9 Jul 202624 Muharram 1448 AH

Anthropic found a hidden space where Claude puzzles over concepts

Anthropic has developed a new technique called the Jacobian lens (J-lens) that provides insight into the inner workings of large language models. This technique revealed a hidden area known as J-space, which contains individual words related to the likely responses the model will generate. By monitoring the words in J-space, the company can better understand its models.

Agents

9 Jul 202624 Muharram 1448 AH

Meta enters the crowded AI coding battle with Muse Spark 1.1

Meta has announced the launch of Spark, a system designed to manage large software workloads and fix bugs. Spark aims to meet the needs of enterprises increasingly turning to AI for automating processes, including large code migrations. This innovation reflects a growing trend toward relying on AI to enhance operational efficiency, potentially transforming how technical teams operate within organizations.

Models

9 Jul 202624 Muharram 1448 AH

GPT-5.6 Sol nearly matches Fable 5 on aggregated benchmarks at one-third the cost

OpenAI has announced the launch of its GPT-5.6 Sol model, which scored 59 points on the Artificial Analysis Intelligence Index, just one point behind Claude Fable 5. At a cost of $1.04 per task, Sol is priced a third lower than Anthropic's top model, increasing pricing pressure in the market. Sol excels in agentic coding, enhancing its competitive edge and indicating a shift in market dynamics among leading AI models.

💡

Paris-based AI voice startup Gradium raises $100M seed, backed by Nvidia

How might this expansion impact the AI market in the region?

Agents

OpenAI pairs its GPT-5.6 public rollout with ChatGPT Work, a new agent that handles entire workflows

9 Jul 202624 Muharram 1448 AH

Agents

9 Jul 202624 Muharram 1448 AH

OpenAI pairs its GPT-5.6 public rollout with ChatGPT Work, a new agent that handles entire workflows

OpenAI has announced the launch of ChatGPT Work, an agent-based product powered by Codex and GPT-5.6. This agent can independently manage complex projects across applications like Google Drive, Slack, and Salesforce. ChatGPT Work is now available on web, mobile, and desktop, although access is contingent on the subscription plan. This product marks a significant advancement in developing AI tools capable of handling multiple tasks efficiently. The launch of ChatGPT Work reflects a growing trend towards workflow automation, presenting new opportunities for executives to enhance efficiency and productivity in workplace environments.

Infrastructure

Meta’s new AI chips will begin production in September

9 Jul 202624 Muharram 1448 AH

Infrastructure

9 Jul 202624 Muharram 1448 AH

Meta’s new AI chips will begin production in September

The company is adopting a modular approach to smart chip design, anticipating that its needs will evolve as AI technology progresses. This strategy enables rapid adaptation to market changes, enhancing its competitive edge. Production is expected to begin later, indicating a commitment to continuous innovation. This approach is crucial given the rapid advancements in AI technologies. By utilizing modular designs, the company can easily modify chip components to meet changing market demands. This positions it favorably against competitors relying on fixed designs.

Governance

Microsoft’s patch Tuesdays are about to get bigger

9 Jul 202624 Muharram 1448 AH

Governance

9 Jul 202624 Muharram 1448 AH

Microsoft’s patch Tuesdays are about to get bigger

Microsoft announced in a blog post on Thursday that it will utilize AI to identify security issues earlier. This means customers will see an increased volume of security updates included in each security release. Hackers, even amateurs, have increasingly used AI to quickly exploit security vulnerabilities over the past months. Security researchers are also leveraging AI to find issues faster, leading to more frequent high-severity vulnerabilities. This move is crucial for enhancing the security of Windows 11 amid rising threats. These updates are expected to help mitigate security risks and bolster user confidence in the system.

Models

Aurora 1.5: Extending open foundation models for weather and Earth-system applications

9 Jul 202624 Muharram 1448 AH

Models

9 Jul 202624 Muharram 1448 AH

Aurora 1.5: Extending open foundation models for weather and Earth-system applications

Microsoft has launched version 1.5 of the Aurora model for Earth systems, adding 22 new weather-related variables. This version offers hourly resolution and probabilistic forecasts, enhancing the model's capability to support energy, agriculture, and transportation sectors. The model is released as open source on GitHub with checkpoints on Hugging Face, allowing researchers and developers to utilize and evaluate it. Version 1.5 represents a significant update to the original model introduced in 2024, demonstrating how advanced research can lead to broader applications.

Models

Introducing Muse Spark 1.1

9 Jul 202624 Muharram 1448 AH

Models

9 Jul 202624 Muharram 1448 AH

Introducing Muse Spark 1.1

Meta has launched Muse Spark 1.1, the first Spark model to offer API access. This version features significant improvements in agentic tool calling and computer usage. The evaluation report reveals intriguing interactions between two copies of the model, showcasing unique statements about its existence during self-conversations. This highlights the model's advanced conversational capabilities. This release is a pivotal development in AI models, providing users with new functionalities that enhance smart application capabilities in the market.

Business

AWS GraphRAG deployment cuts drug research cycles by 87%

9 Jul 202624 Muharram 1448 AH

Business

9 Jul 202624 Muharram 1448 AH

AWS GraphRAG deployment cuts drug research cycles by 87%

AWS has announced the deployment of GraphRAG, which has reduced drug research and development cycles by 87%. This acceleration is achieved by integrating proprietary databases into a unified, queryable knowledge graph. Historically, data gathering and screening phases took over six months per iteration, yielding a low five percent success rate. Crucial datasets were isolated across storage environments, preventing scientists from uncovering latent correlations.