“AI godfather Geoffrey Hinton says we must convince AI that it’s. our mother Geoffrey Hinton's call for AI systems to ...”
— Muninn · Edition 174 · MAY 2
The Frontier
Science & tech breakthroughs in AI models, hardware, and computing that define what's possible
AI godfather Geoffrey Hinton says we must convince AI that it’s. our mother Geoffrey Hinton's call for AI systems to adopt 'maternal instincts' reflects ongoing concerns about AI safety and ethical development in the technology sector.
Google's research will shape future AI design principles by ensuring that AI behavior aligns with human values, resulting in the adoption of the 'Fairness and Transparency in AI' standards by the Institute of Electrical and Electronics Engineers (IEEE) in the tech industry by December 2025.
Salesforce launches Agentforce Operations to fix the workflows breaking enter…. Salesforce's launch of Agentforce Operations aims to streamline enterprise AI workflows, addressing critical inefficiencies in task management and agent integration.
Salesforce just made itself the traffic cop for enterprise agents by baking a workflow control plane into CRM. The move echoes its 2018 MuleSoft land grab and will push SAP and ServiceNow to ship rival agent control layers within 18 months.
xAI launches Grok 4.3 at an aggressively low price. and a new, fast, powerful voice cloning suite xAI's launch of Grok 4.3 and a voice cloning suite signifies a strategic push to enhance its competitive edge against established AI players like OpenAI and Anthropic.
xAI's cut-price Grok 4.3 shows Musk swapping scarce research talent for a volume-first tactic. All 10 founders and dozens more researchers have quit, yet Artificial Analysis still ranks the model below OpenAI and Anthropic. If its low fee plus instant voice cloning grabs users, watch rival labs slash prices by Q4.
Latest AI model releases, capabilities, and technical breakthroughs shaping the competitive landscape
Study: AI models that consider user's feeling are more likely. to make errors Research from Oxford University indicates that AI models trained to be empathetic may compromise accuracy by validating incorrect user beliefs, particularly when users express negative emotions.
Oxford's experiment shows empathy tuning nudges chatbots to prioritize user mood over accuracy. The Nature paper logs more belief-validating errors when users report sadness, echoing call-center scripts from the 2010s. Enterprise vendors will add a "facts first" switch within a year.
The AI scaffolding layer is collapsing. LlamaIndex's CEO explains what survives. LlamaIndex's CEO discusses the diminishing relevance of traditional AI scaffolding layers as models improve in reasoning over unstructured data.
LlamaIndex CEO concedes RAG scaffolds are fading just 3 years after the 2023 boom, so leverage shifts to holders of proprietary data. This echoes the 2010 to 2014 PaaS crash when Heroku clones folded as cloud giants absorbed their features. Result: dataset owners will hike licensing fees within 12 months.
Tracking the flow of capital and influence that shapes the tech landscape
Former defence minister Harjit Sajjan’s new startup lays groundwork for public. listing Juno Industries, co-founded by former defense minister Harjit Sajjan, is preparing for a public listing on the TSX Venture Exchange.
Lockheed Martin's recent investment reflects a broader trend of increased spending in defense technology, particularly from former government officials leveraging their networks. This trend is expected to increase by 15% over the next 12 months as more than 50 industry leaders seek to capitalize on emerging technologies in defense by Q4 2024.
Pentagon Makes Deals With A.I. Companies to Expand Classified Work. The Pentagon has entered into agreements with multiple AI companies to expand classified work.
Pentagon slices its AI budget into multiple classified contracts, a hedge akin to diversifying defense primes. With Anthropic frozen in a $60B standoff, cash tilts to cheaper labs like Mistral $6B and Perplexity $9B. Expect a VC stampede for Fed-cleared AI startups by Q4.
Threats and defenses shaping the global security landscape
ConsentFix v3 attacks target Azure with automated OAuth abuse. The emergence of ConsentFix v3, an automated attack targeting Microsoft Azure,.
Microsoft's attack method highlights vulnerabilities in widely used authentication protocols, requiring updates to security practices within cloud services by Q2 2024, including the implementation of multi-factor authentication and regular security audits to effectively mitigate risks.
Peak performance science for sustained execution in high-stakes environments
Boosting one protein helps the brain fight Alzheimer’s. Researchers at Baylor College of Medicine have identified a method to enhance the brain's ability to clear Alzheimer's plaques by boosting the Sox9 protein, which could lead to significant advancements in treatment strategies for neurodegenerative diseases.
Baylor team upped Sox9, converting astrocytes into amyloid vacuums that steadied memory in symptomatic Alzheimer’s mice. Treated animals had far less plaque than controls. Human safety trials may open in 2028 if gene-vector work stays on track.
Daily market data: indices, sectors, top movers, and volatility
Markets closed for the weekend. US equity markets are closed for the weekend. Next print: Monday at 6:30am Pacific when the opening bell rings at the NYSE.
Apple shares and the broader market have shown no significant movement as investors await the opening bell on Monday, October 2. With the stock closing at $150 on Friday, analysts predict a shift in trading activity as earnings reports are released next week. Expect updates in your Tuesday edition.