AI News

Follow the latest AI and automation news

4 Jul 202619 Muharram 1448 AH

Alibaba DAMO Academy's ElementsClaw AI Agent Discovers 4 New Superconductors in Just 28 GPU Hours

Alibaba's DAMO Academy, in collaboration with Renmin University, has launched ElementsClaw, the first AI agent dedicated to discovering superconducting materials. The system analyzed 2.4 million stable crystal structures in just 28 hours using GPUs, identifying 68,000 superconductors and experimentally verifying four previously unknown ones. Since the first superconductor was discovered in 1911, around 2,000 superconducting materials have been recognized, with fewer than 100 exhibiting superconductivity above 40 Kelvin.

Models

3 Jul 202618 Muharram 1448 AH

Open Source AI Gap Map

Current AI has launched the Open Source AI Gap Map, detailing 421 products. This includes 266 software tools, 85 models, 50 datasets, and 20 hardware projects from 228 organizations, organized into 14 categories across three layers of the stack. The map aims to index the current state of open-source AI, with 24,400 uncategorized artifacts awaiting research and citation. This structured approach highlights the gaps in the ecosystem and encourages further exploration of unclassified resources.

Agents

3 Jul 202618 Muharram 1448 AH

Microsoft follows Anthropic and OpenAI into the AI super app race with overhauled Copilot and AutoPilot agents

Microsoft plans to merge its consumer and enterprise Copilot apps into a single application by August. Features like Copilot Podcasts will be discontinued, while new AI agents called 'AutoPilot' will manage tasks in the background for an additional fee. This move aligns with Microsoft's strategy to compete in the smart app market, aiming to provide a seamless user experience. The integration of apps will simplify operations and enhance the effectiveness of AI tools in business environments.

💡

Security vulnerability reports have exploded since AI models started hunting for bugs

What strategies can be implemented to enhance security in light of these findings?

Models

UK's AI Security Institute finds standard benchmarks systematically underestimate what AI agents can actually do

3 Jul 202618 Muharram 1448 AH

Models

3 Jul 202618 Muharram 1448 AH

UK's AI Security Institute finds standard benchmarks systematically underestimate what AI agents can actually do

The UK's AI Security Institute found that standard AI evaluations underestimate agent capabilities by limiting the compute budget. Success rates in software engineering tasks increased by about 25% when the token budget was raised tenfold. Newer models benefit significantly from this increase. The findings suggest that actual progress at the frontier is about 60% steeper than previous measurements indicated, highlighting the need to reassess current evaluation standards. This reflects how compute budget constraints can skew our understanding of AI capabilities.

Agents

Tesla Launches Robotaxi Service in Miami to Boost Self-Driving

3 Jul 202618 Muharram 1448 AH

Agents

3 Jul 202618 Muharram 1448 AH

Tesla Launches Robotaxi Service in Miami to Boost Self-Driving

Tesla has officially launched its robotaxi service in Miami, enhancing its efforts to expand autonomous transportation. The service is available to users starting today and is part of Elon Musk's strategy to transform Tesla into an AI-focused entity. Tesla faces intense competition from companies like Waymo and Zoox, making this expansion increasingly significant. The service was first launched in Austin, Texas, in June, with plans to expand to Dallas and Houston. Elon Musk anticipates that self-driving cars will become more prevalent on American roads by the end of the year.

Agents

Architectural Guide To Error Handling for LLM Tool Calling

3 Jul 202618 Muharram 1448 AH

Agents

3 Jul 202618 Muharram 1448 AH

Architectural Guide To Error Handling for LLM Tool Calling

Handling errors in AI tool calls in production environments is crucial to prevent automated systems from failing when a connected service encounters issues. The article outlines multi-layered defense strategies that include failure types, retry strategies, and fallbacks. Classifying errors in tool calls is essential for system continuity. The system should identify the cause of failure rather than blindly submitting requests or throwing a general exception. This operational logic necessitates dividing recovery responsibilities between two layers: the coordination layer and the model itself.

Models

Anthropic wants to develop its own drugs

3 Jul 202618 Muharram 1448 AH

Models

3 Jul 202618 Muharram 1448 AH

Anthropic wants to develop its own drugs

At the recent event 'The Briefing: AI for Science', Anthropic unveiled Claude Science, an AI workbench aimed at scientists. This platform integrates fragmented tools and datasets, enabling users to generate visuals and figures efficiently. Anthropic, a leader in AI with its advanced coding tools, emphasizes that Claude Science can significantly accelerate scientific discovery and healthcare interventions. The company highlighted its existing partnerships with various biotech and pharma clients already utilizing this tool.

Models

GPT and Claude failed Bridgewater's finance tests because the right answers were never public

3 Jul 202618 Muharram 1448 AH

Models

3 Jul 202618 Muharram 1448 AH

GPT and Claude failed Bridgewater's finance tests because the right answers were never public

Bridgewater and Thinking Machines Lab report that a finely tuned open-weight model outperforms leading AI models in evaluating financial documents. Their analysis shows this model achieves superior results at a significantly lower cost. The findings reveal that models like GPT and Claude failed Bridgewater's finance tests due to the lack of publicly available correct answers. This highlights the critical role of accessible data in assessing model performance.

Business

Takeda signs US$600M AI drug discovery deal with Insilico

3 Jul 202618 Muharram 1448 AH

Business

3 Jul 202618 Muharram 1448 AH

Takeda signs US$600M AI drug discovery deal with Insilico

Takeda has announced a strategic collaboration with Hong Kong's Insilico Medicine to leverage AI in drug discovery across its therapeutic areas. The specific therapeutic areas and disease targets remain undisclosed, but the agreement provides Takeda access to Insilico's Pharma.AI platform. The collaboration aims to identify drug candidates that meet predefined scientific criteria, with Insilico leading the AI-driven discovery efforts. The deal includes approximately $60 million in project initiation fees, potentially reaching $600 million with successful milestones.