 Amazon releases Strands Decider, a model similar to TypeSafe's JevAmazon Web Services has launched Strands Decider 2B, an open-source model that delivers faster decisions at a lower cost compared to large language models for agentic workflows. This model is positioned as an AI tool that can be run locally, offering confidence scores by selecting among predefined options. |
AiPulse — Issue No. 75
COVER
Pulse
The company stated that these individuals, who are alleged to have leaked confidential information to a third party, violated its policies. The incident occurred immediately after OpenAI canceled the launch of the GPT-6.1 Astra model due to security concerns. |
At Work
A tool developed by researchers from MIT CSAIL, Google, and Northeastern University allows users to easily edit designs that AI has generated but that are non-functional. As a result, both experts and beginners can produce 3D-printable and functional objects. |
The workshop teaches participants, through hands-on practice, AI-assisted idea development and editable 3D modeling processes. The event will be held online on November 15, 2026. |
Verifiable capability architecture proposed for autonomous laboratoriesResearchers at Nanjing Medical University proposed a three-layer architecture that enables autonomous laboratories to provide instant proof of competence for each task. This system verifies that devices can safely perform the task under current conditions using machine-readable records that include uncertainty information. |
Global Lens
Microsoft has released MAI-Transcribe-2-Streaming, which offers real-time transcription, and new speech synthesis models as a public preview on Microsoft Foundry. These models are designed for voice assistants and interactive applications that operate with low latency. |
According to Epoch AI data, AI performance costs are falling faster than DNA sequencing and lithium batteries. This rapid price erosion is challenging the profitability and customer loyalty targets of companies like OpenAI and Anthropic. |
The U.S. Federal Trade Commission has launched a formal investigation to examine the consumer-facing risks of artificial intelligence laboratories. This step was taken just one day after the companies signed a voluntary security agreement at the White House. |
Frontier
Hermes: Learning Contextual Reasoning Unlocks Test-Time ScalingThe research shows that the ability of models to effectively distribute computational power across context windows during inference is defined as 'contextual reasoning' and that this skill is learnable. Models trained with the Hermes-Learn framework improve their performance by better leveraging additional computational power. |
The system analyzes the target model's failures to create new tasks focused on deficient competencies. This approach improves the training process to ensure agents operate more reliably in enterprise environments. |
Watch & Listen
Zhang argues that taking on high-risk research bets that appear "trivial" but are too risky for industrial laboratories to undertake has yielded critical advances, particularly in areas such as RLMs and multi-agent systems. This perspective highlights the potential of academic circles to produce innovative solutions even with limited resources. |
Fun & Curious
American researchers, in collaboration with Berkeley Lab and industry partners, are working on new EUV lithography materials that will enable the production of chips that are a few atoms smaller and faster. This development is of critical importance for maintaining access to the most advanced chips in terms of economic security and national defense. |
FOLLOW
Get new issues in your feed reader — add this to an app such as Feedly, Inoreader or NetNewsWire. Follow by RSS |