The White House AI task force is set to hold its first meeting this week, though pressure is mounting from members of Congress to establish enforceable safety standards. Senators Mike Rounds and Maria Cantwell are pushing for more resources at the National Institute of Standards and Technology. The group, led by AI czar Jay Clayton, has 120 days to deliver a report on AI risks covering biosecurity, cybersecurity, and threats to critical infrastructure, financial services, energy, transportation, and emergency services.
Amazon is in advanced talks to acquire AI startup Decart for $7 billion, a move that would integrate Decart's workload optimization software into Amazon's custom silicon stack, including Trainium. Aidan Marcuss is expected to lead Amazon's Alexa division as the company seeks to convert its large AI-focused capital expenditure into long-term revenue. The deal underscores how major cloud providers are racing to strengthen their AI infrastructure and voice assistant capabilities.
Nvidia CEO Jensen Huang spoke in Paris on October 13, calling the AI era an opportunity for Europe to catch up with other regions. Nvidia announced minority investments in four French startups working in healthcare, materials, robotics, and business decision support. Huang leads the world's most valuable company, with a market value of $5.47 trillion.
On the fraud detection front, EXL demonstrated that adding graph context to AI models improved fraud capture by 27 to 33 percent. The model uses neighborhood data to score transactions, turning a 0.04 score into a 0.91 score when linked entities are included, while maintaining P99 latency at 3.1 milliseconds. Separately, Apate.ai deploys AI agents that pose as humans to interact with scammers, wasting 17 years of scammers' time across 2.5 million conversations. Its customers include banks, telecom companies, and government agencies.
AI trading agents face strict reliability demands because they chain multiple calls together, meaning one failure can break an entire workflow. Binance reported 90,000 daily requests with a 97% success rate and 95% of responses under 60 milliseconds in its first two weeks of operation. Experts say reliability must be built into these systems from the start. Meanwhile, a new study shows that Evidence-Centered Design improves prompt engineering for GenAI models, outperforming state-of-the-art prompting baselines in automated essay scoring.
Key Takeaways
- The White House AI task force, chaired by AI czar Jay Clayton, has 120 days to deliver a report on AI risks including biosecurity, cybersecurity, and critical infrastructure threats.
- Amazon is in advanced talks to acquire AI startup Decart for $7 billion to bolster its custom silicon stack, including Trainium.
- Aidan Marcuss is set to lead Amazon's Alexa division as the company pivots its AI capital spending into long-term revenue.
- Nvidia CEO Jensen Huang announced minority investments in four French startups across healthcare, materials, robotics, and business decision support.
- Nvidia holds a market value of $5.47 trillion, making it the world's most valuable company.
- EXL's graph-enhanced AI models improved fraud capture by 27 to 33 percent while holding P99 latency at 3.1 milliseconds.
- Apate.ai's undercover AI agents have wasted 17 years of scammers' time across 2.5 million conversations, serving banks, telecoms, and government agencies.
- Binance's AI trading agents achieved a 97% success rate across 90,000 daily requests, with 95% of responses under 60 milliseconds.
- Evidence-Centered Design outperforms existing prompting baselines for GenAI models in automated essay scoring.
- Congressional pressure is growing for enforceable AI safety standards, with senators pushing for more NIST resources.
White House AI task force plans first meeting amid pressure
The White House AI task force is preparing for its first meeting this week. Groups and members of Congress are pushing the body to create enforceable safety standards for AI. Senators Mike Rounds and Maria Cantwell want more resources for the National Institute of Standards and Technology. The FBI also arrested a Saudi national this week. No one knows exactly what will be discussed yet.
Trump's Super Intelligence Force gets its leader and goals
President Trump created the Super Intelligence Force to keep America leading in AI development. Jay Clayton, named AI czar in October, will chair the task force. The group has 120 days to deliver a report on AI risks. The reported charter focuses on biosecurity, cybersecurity, and threats to critical infrastructure. It also includes financial services, energy, transportation, and emergency services.
EXL graph model boosts fraud detection with AI
EXL showed that adding graph context to AI models improved fraud capture by 27 to 33 percent. The company presented its findings at AI Engineer. The model uses neighborhood data to score transactions, turning a 0.04 score into a 0.91 score when linked entities are included. The system holds P99 latency at 3.1 milliseconds. EXL runs two separate models for physical point-of-sale fraud and online fraud.
Amazon in talks to buy Decart for 7 billion dollars
Amazon.com is in advanced talks to acquire AI startup Decart for 7 billion dollars. The deal would add Decart's software for optimizing AI workloads into Amazon's custom silicon stack, including Trainium. Aidan Marcuss is set to lead Amazon's Alexa division. The move reflects a broader shift in Amazon's AI infrastructure and voice assistant plans. Amazon wants to turn its large AI-focused CapEx program into long-term revenue.
Evidence-Centered Design improves AI prompt engineering
A new study proposes using Evidence-Centered Design to guide prompt engineering for GenAI models. The framework maps student, evidence, and task models onto prompt design. Researchers found that structured prompting strengthens validity and reliability of AI-generated educational outputs. The study focused on automated essay scoring. Results showed ECD-designed prompts outperformed state-of-the-art prompting baselines.
AI trading agents must meet strict reliability standards
AI trading agents need higher reliability than regular apps because they make many fast decisions. Binance reported 90,000 daily requests with a 97% success rate and 95% of responses under 60 milliseconds in the first two weeks. Even small failure rates matter because agents chain multiple calls together. One failed step can break an entire trading workflow. Experts say reliability must be built into the system from the start.
Nvidia CEO Jensen Huang shares optimism about AI future
Nvidia CEO Jensen Huang spoke in Paris on October 13 about the future of AI. He called the AI era a chance for Europe to catch up with other regions. Nvidia also announced minority investments in four French startups. These companies work in healthcare, materials, robotics, and business decision support. Huang leads the world's most valuable company with a market value of $5.47 trillion.
Apate uses AI agents to fight scams from the inside
Apate.ai sends AI agents undercover to interact with scammers and collect fraud data. Founded in 2024 based on research from Macquarie University in Sydney, the company uses agents that answer scam calls and click phishing links. The agents have wasted 17 years of scammers' time across 2.5 million conversations. Customers include banks, telecom companies, and government agencies. The company shares stolen details like bank accounts and crypto wallets to help prevent fraud.
Nigerian designer Adekunle expands AI and fintech work
Oladosu Ibrahim Adekunle is a Nigerian-born product designer based in the United Kingdom. He currently works at Trussell's Camden Foodbank designing tools for foodbank operations. Previously he led UX design at Xenet AI for Yodoo, an AI chat platform. He also worked on digital banking tools for Zenith Bank and other institutions. His goal is to build globally competitive products that highlight African talent in technology.
New method improves pattern learning for ranking systems
This research addresses how to rank data patterns based on user preferences. The team uses Analytic Hierarchy Process to weigh different measures according to what users value. They also use active learning with a sensitivity-based heuristic to reduce the number of ranking queries needed. Tests show the method cuts running time while delivering precise results. The approach is also robust to user error compared to existing techniques.
Sources
- AI task force readies for first meeting
- What is the US Super Intelligence Force and who leads it?
- EXL's graph model finds fraud in the edges
- What Could Amazon.com (AMZN) Gain From A $7 Billion AI Deal And Alexa Shift?
- Using Evidence-Centered Design to Guide Prompt Engineering in Measurement: An Illustration with Automated Essay Scoring
- AI trading agents face a reliability bar human apps never did
- Nvidia CEO Jensen Huang 'optimistic' on AI risks, sovereignty and jobs
- Apate sends AI agents undercover to get scammed so they can gather fraud intelligence
- UK-based Nigerian product designer, Adekunle expands portfolio across AI, Fintech, international markets
- Boosting the Learning for Ranking Patterns
Comments
Please log in to post a comment.