Advancing AI Development

Did Stanford just kill LLM fine-tuning? . . This new paper from Stanford, called Agentic Context Engineering (ACE), proves something wild: you can make models smarter without changing a single weight. Here's how it works: Instead of retraining the model, ACE evolves the context itself. The model writes its own prompt, reflects on what worked and what didn't, then rewrites it. Over and over. It becomes a self-improving system. Think of it like the model keeping a living notebook where every failure becomes a lesson and every success becomes a rule. The results are impressive: - 10.6% better than GPT-4-powered agents on AppWorld - 8.6% improvement on financial reasoning tasks - 86.9% lower cost and latency No labeled data required. Just feedback loops. Here's the counterintuitive part: Everyone's chasing short, clean prompts. ACE does the opposite. It builds dense, evolving playbooks that compound over time. Turns out LLMs don't need simplicity. They need context density. The question here is how to manage all this information and experience. This is where building a real-time memory layer for Agents like Zep AI (YC W24) can be a great solution and active area of research going forward. What are your thoughts? I have linked the paper in the next tweet! ____ If you found it insightful, reshare with your network. Find me → Akshay Pachaar ✔️ For more insights and tutorials on LLMs, AI Agents, and Machine Learning!

Some more big news for Canadian AI 🍁 The Government of Canada has signed an MOU with Cohere to explore how AI could be applied in public services. This comes on the heels of Ottawa’s earlier $240M investment. Why does this investment matter, and what does it signal?  • Shows the government’s willingness to actually test Canadian-built AI in real use cases  • It could open procurement doors that accelerate adoption (we know this is desperately needed in Canada)  • Puts Canada in the mix with peers like the UK that are already testing sovereign AI approaches. Beyond this, Canadian AI is having a real moment— Blue J, GeologicAI, Clio, and Waabi have all seen big funding or acquisitions recently. The challenge? Making sure Canada doesn’t just grow AI companies for someone else to buy, but keeps capacity here at home. If this MOU turns into real contracts, it could mark a tipping point for Canadian AI sovereignty and adoption. If it doesn’t..... well, let’s just say it’ll look great in a press release. Either way, it’s a sign Canada is serious about building with Canadian AI. #AI #Canada #GovTech

We just released the world's first paper on Quantum Agentic AI, focusing on the new, LLM-driven notion of agentic AI that currently transforms how we build autonomous agents! Our team including Prof Bill Buchanan OBE FRSE, Dr. Mark Tehrani, Muhammad Shahbaz Khan and Siddhant Dutta dove deep into the intersection of quantum computing and the new, LLM-driven notion of agentic AI—the concept of autonomous agents that leverage large language models to plan, decide, and act intelligently. Key highlights of our work are: ● The first formal definition of Quantum Agents based on LLM-inspired agentic principles ● New architectures that tightly integrate quantum processors with agent-based reasoning ● Three working prototypes: from Grover-based decision-making to adaptive quantum encryption ● Use cases spanning quantum-enhanced edge AI, chemistry, defense, and hybrid optimization What is the amazing and relevant big thing? Agentic AI is redefining autonomy and decision-making. By bringing quantum computing into the loop, we unlock entirely new horizons—systems that combine the best of classical and quantum intelligence. We’d love to hear your thoughts: Where do you see the biggest impact of Quantum Agents? What real-world problems could they solve today? Wir haben das weltweit erste Paper zu Quantum Agentic AI veröffentlicht, und zwar mit einem Fokus auf den neuen, von LLMs geprägten Agentic-Begriff, der aktuell die Entwicklung von autonomen Agenten neu definiert! Unser Team hat sich an der Schnittstelle von Quantencomputing und dem neuen, von LLMs geprägten Agentic-Begriff positioniert, also Agenten, die mit großen Sprachmodellen eigenständig planen, entscheiden und handeln. Highlights aus unserer Arbeit: ● Die erste formale Definition von Quantum Agents, basierend auf LLM-inspirierten agentischen Prinzipien ● Neue Architekturen, die Quantenprozessoren nahtlos mit Agenten-Logik verknüpfen ● Drei funktionierende Prototypen: vom Grover-basierten Entscheidungsagenten bis zur adaptiven Quantum Image Encryption ● Anwendungsfelder von Quanten-Edge-AI über Chemie bis Verteidigung und hybride Optimierung Was ist das Geniale daran? KI-Agenten verändern derzeit, wie wir über Autonomie und Entscheidungsfindung denken. Mit Quantum Agents kombinieren wir diese neue Form von Intelligenz mit den Fähigkeiten des Quantencomputings – für Anwendungen, die bisher undenkbar waren. Deine Meinung interessiert uns: Wo siehst du die größten Potenziale für Quantum Agents? Welche Probleme könnten Quantum Agents schon heute lösen? #QuantumAI #AgenticAI #QuantumComputing #LLM #AIResearch #Innovation #FutureOfWork #QuantumAgents

Linkedin was super active yesterday on the back of Prime Minister Trudeau’s pre-budget announcement of $2.4 billion for #AI related investments! This is a major step forward, and worthy of enthusiasm! Measures mentioned include:   ·     $2 billion to fund access to computing capabilities and technological infrastructure – including a new AI #compute access fund and development of a Canadian AI Sovereign Compute Strategy. ·     $200 million to advance AI #startups in sectors such as agriculture, clean technologies, healthcare, and manufacturing. ·     $100 million in the NRC IRAP AI assist program for SMEs to scale up productivity with AI solutions. ·     $50 million for the Sectoral Workforce Solutions Program, providing new skills training for workers in disrupted sectors and communities. ·     $50 million towards a new Canadian Safety Institute. ·     $5.1 million to strengthen enforcement of the proposed AI data Act.   That said, although all measures are welcomed, two parts of the plan jump out to me as particularly interesting:   Firstly, the infrastructure investment. This is amazing news and an absolutely necessary investment. Limited access to compute stymies the social, environmental, and economic potential sitting inside Canada’s talented AI ecosystem and is a problem the government has to play a role in solving. Admittedly, this investment is a first step to levelling of the playing field as most large economies have already announced significant investment on this front. The proportion of investment into developing sovereign compute capacity and the response of private markets will be interesting to see. In this context, the national and international investments made by US-based cloud providers have offered them an incredible lead in infrastructure which amounts to a tremendous accumulation of AI power for the US. It is likely that Canada will have to share AI power rather than outrightly target AI autonomy, but the way this unfolds will influence much of Canada’s AI leadership quest.   Secondly, the $50 million investment into a new Canadian Safety Institute. This is also a levelling of the playing field with the US and UK already establishing Safety advisory groups and institutes and securing cooperation with OpenAI, Meta, and Nvidia. That being said, although Canadians tend to bemoan our risk aversion as a hindrance to technological progress, it can be a massive strength in the advancement of the AI. The Canadian Safety Institute has the potential to demonstrate Canadian leadership at the intersection of value based and economic objectives. I believe that our “Canadian-ness” can and will be a significant advantage.   Overall, these are directionally important updates and provide much needed momentum to a promising future with AI at the core. My sincere hope is that those Canadian AI innovators who work on solving the world’s most significant challenges ultimately accrue most of the value from this and future investments.

I've heard so much noise lately about AI adoption in Canadian businesses. I've heard that entrepreneurs aren't ambitious enough. I've heard that we're laggards. I've heard that small businesses are risk-averse. I read a particularly offensive line that "many Canadian businesses never miss an opportunity to miss an opportunity." Um, I don't think so. The small and medium-sized enterprises (SMEs) I work with every day aren't risk-averse. They don't lack ambition. They've taken some serious punches in the past few years and are still kickin'. And when it comes to their AI adoption, they're not missing an opportunity. The SMEs I know are eager to adopt AI, they just need some guidance on where to begin, and advice on strategy. As Kirsten Koppang Telford of The Forum, Sarah Stockdale of Growclass, and I say in our op ed in today's edition of The Hill Times: All Canadians need access to AI training, but business owners need more than just a course. They need support systems that work in practice. That’s what turns training into lasting organizational change. This isn’t the time for passive optimism. AI alone won’t solve our productivity problem but people will, if we give them the resources they need. As co-leads of the AI Skills Lab Canada pilot, with co-investment by DIGITAL, Canada’s Global Innovation Cluster for digital technologies, we launched the country’s first program to help women and non-binary entrepreneurs adopt AI. Since April, the AI Skills Lab Canada pilot has trained 103 women and non-binary entrepreneurs and business leaders using a wayfinding approach with expert-led instruction, small peer-learning cohorts, practical AI integration roadmaps, and support from AI coaches. And it’s working. Participants’ ability to set up AI systems and processes grew by 90 per cent, and confidence in selecting AI tools increased by 89 per cent. Their understanding of ethical and regulatory considerations rose by 119 per cent. When AI training is timely, practical, and supported by a trusted peer network, people apply what they learned. That’s not just a win for inclusion. It’s a win for the economy. When entrepreneurs have the tools and training to adopt AI in a way that is values-aligned, more businesses can grow, hire, and innovate. Equity and productivity move together. Now is the moment for Canada to be as ambitious about equitable AI adoption as we are about AI innovation. Let's not waste it. (link to the op ed in The Hill Times is in the comments below)

Large language models (LLMs) can improve their performance not just by retraining but by continuously evolving their understanding through context, as shown by the Agentic Context Engineering (ACE) framework. Consider a procurement team using an AI assistant to manage supplier evaluations. Instead of repeatedly inputting the same guidelines or losing specific insights, ACE helps the AI remember and refine past supplier performance metrics, negotiation strategies, and risk factors over time. This evolving “context playbook” allows the AI to provide more accurate supplier recommendations, anticipate potential disruptions, and adapt procurement strategies dynamically. In supply chain planning, ACE enables the AI to accumulate domain-specific rules about inventory policies, lead times, and demand patterns, improving forecast accuracy and decision-making as new data and insights become available. This approach results in up to 17% higher accuracy in agent tasks and reduces adaptation costs and time by more than 80%. It also supports self-improvement through feedback like execution outcomes or supply chain KPIs, without requiring labeled data. By modularizing the process—generating suggestions, reflecting on results, and curating updates—ACE builds robust, scalable AI tools that continuously learn and adapt to complex business environments. #AI #SupplyChain #Procurement #LLM #ContextEngineering #BusinessIntelligence

For years, fine-tuning LLMs has required large amounts of data and human oversight. Small improvements can disrupt existing systems, requiring humans to go through and flag errors in order to fit the model to pre-existing workflows. This might work for smaller use cases, but it is clearly unsustainable at scale. However, recent research suggests that everything may be about to change. I have been particularly excited about two papers from Anthropic and Massachusetts Institute of Technology, which propose new methods that enable LLMs to reflect on their own outputs and refine performance without waiting for humans. Instead of passively waiting for correction, these models create an internal feedback loop, learning from their own reasoning in a way that could match, or even exceed, traditional supervised training in certain tasks. If these approaches mature, they could fundamentally reshape enterprise AI adoption. From chatbots that continually adjust their tone to better serve customers to research assistants that independently refine complex analyses, the potential applications are vast. In today’s AI Atlas, I explore how these breakthroughs work, where they could make the most immediate impact, and what limitations we still need to overcome.

A Ukrainian operator compared it to a video game: set the waypoints, pick the targets, and let it run. He was talking about a drone mothership that flies 300 kilometers, drops two AI-guided FPVs, and returns home—no comms, GPS, or pilot. According to Strategy Force Solutions, they’ve already used the system in live trials against Russian targets. It’s unconfirmed, but credible. And it’s exactly the kind of autonomy the defense world has been theorizing for years. What’s striking isn’t the drone itself, it’s the software stack behind it. A LIDAR-based autonomy suite originally built for civilian infrastructure inspection, now retooled for war. The drone sees, navigates, and strikes the way a human would, but faster, with fewer constraints, and no need for a remote operator. This capability has grown essential as the battlefield has evolved. Jamming and electronic warfare have made the skies above Ukraine chaotic for traditionally-controlled drones, but the country's military has adapted in two distinct ways: looking backward to fiber-optics, and forward to edge-deployed autonomy. The latter unlocks resilience—drones that don’t need to phone home, that can make decisions on their own, and complete missions even in contested, comms-denied environments. If it works, it’s not just another edge case. It’s a glimpse at where this is all heading: kill chains designed around AI-first logic, not human workflows. And the most important part? It’s already flying. Built under siege. Fielded at scale. We keep asking what autonomy can augment. But we’re past that. The better question now: what happens when autonomy is the force?

Eight years ago, we set out to build an AI ecosystem that could compete globally. New Vector Institute research prepared by Deloitte Canada proves that vision is reality—and the numbers are remarkable. AI-related jobs have contributed between $82 billion and $100 billion to Canada's economy over the past five years, with Ontario accounting for nearly half of this total. But as Deloitte Canada's Chief Economist Dawn Desjardins notes: "The impact of AI is already apparent; we're seeing this firsthand through the establishment of numerous AI labs and the economic impact of AI-related jobs. Our analysis shows that continued adoption of AI across the Canadian economy has the potential to drive significant economic growth, enhance labour productivity, and generate net new jobs—growth that would not be achievable without the integration of AI technologies." Overall, the research reveals compelling evidence of AI’s national and provincial impact: ➡️ Federal investment of $1.1 billion nationwide attracted $10.64 in private sector investment for every public dollar ➡️ Ontario attracted $446 million in federal AI investment, generating $9.53 in private investment for every federal dollar ➡️ Over 17,000 new AI jobs created in Ontario this past year ➡️ Canada projected to achieve $298 billion in AI-driven economic growth over the next decade The foundation is built. The question now is whether we can maintain the momentum to scale. Canada leads the G7 in AI talent growth, but talent follows opportunity. This research proves that strategic public investment catalyzes private sector commitment, creating an ecosystem where breakthrough research translates into companies, jobs, and economic growth. The global competition for AI leadership is intensifying. Our proven ecosystem gives us a distinct advantage—but only if government, industry, and research institutions continue working together. Thank you, Dawn, and the team at Deloitte Canada, including Audrey Ancion, and Anthony Viel, for collaborating with our Vector team members including Craig Stewart, Bob (YiAn) Zhou, to gather this comprehensive analysis, and to the federal and provincial governments whose strategic support continues delivering results.

One of the first papers in the World to outline quantum and agentic AI? This paper explores the intersection of quantum computing and agentic AI by examining how quantum technologies can enhance the capabilities of autonomous agents, and, conversely, how agentic AI can support the advancement of quantum systems. We analyze both directions of this synergy and present conceptual and technical foundations for future quantum-agentic platforms. Our work introduces a formal definition of quantum agents and outlines potential architectures that integrate quantum computing with agent-based systems. As a proof-of-concept, we develop and evaluate three quantum agent prototypes that demonstrate the feasibility of our proposed framework. Furthermore, we discuss use cases from both perspectives, including quantum-enhanced decision-making, quantum planning and optimization, and AI-driven orchestration of quantum workflows. By bridging these fields, we aim to chart a path toward scalable, intelligent, and adaptive quantum-agentic ecosystems. Eldar Gunter Sultanow, Dr. Mark Tehrani, Siddhant Dutta, Muhammad Shahbaz Khan https://lnkd.in/eDDmTWtQ