Our first speaker for today is the dean of Steuart School of Business at Illinois Tech. He's an expert in institutional effectiveness and student success. A graduate of Harvard's education leadership programs and a US Army aviation veteran. He brings a unique perspective on leadership, technology, and education transformation.To kick off our program today, he asked a question that we all have been wondering. Can machines think. So, please give a big round of applause to Dean Rich Clay.Well, good morning and ...

If highbrow maths-literate large language models can certify existing proofs, and help develop new ones, many mathematicians hope they could speed up discovery https://t.co/RAM57Kh17C ...

It's kind of a process of trial and error and I think that's how most people end up using AI. You start by asking a question and when you get the response you realize, oh, I forgot to add this piece of information, that piece of information. So, you have to ask a second, a third, a fifth question.And by the time you're done, you've maybe asked 20 or 30 prompts. Well, there are ways to shortcut that. One way is that after you go through a sequence of these prompts and finally get the answer that really addre ...

🚨NEW: GROK IS GRADUALLY BECOMING THE NEW GOOGLEGrok has officially recorded 293.3 million website visits over the last 30 days, marking a surge in independent web adoption.The AI platform now ranks among the top 70 most-visited websites globally as it expands significantly beyond its initial integration with @X.This rapid growth indicates a shifting competitive landscape for Large Language Models as users increasingly seek alternatives to legacy incumbents. ...

Developers just shipped a new class of AI Agents!To understand why it matters, you need to see where it sits.Level 1: Prompt → ResponseEach call is stateless. The model can use tools/APIs within a single request but nothing persists. Most production LLM apps are sophisticated Level 1 wrappers.Level 2: Interactive assistantThe platform handles persistence for you with memory, tools, files, connectors. ChatGPT and Claude live here. These are capable, but entirely reactive.Level 3: Delegated executionYou defin ...

All software will be self-evolving software very soon. This is a new paper from a team out of Stanford, MIT and Crafted and it's called Meta Harness endtoend optimization of model harnesses. If you've been following the AI industry at all, if you've watched this channel at all recently, I have been talking about aentic harnesses a ton.What is a harness. It is basically the traditional code that is wrapped around a model like Claude, like GPT54, like Gemini that tells it how to operate that allows it to stor ...

Core Insights - The article discusses the transformative impact of large language models (LLMs) on the development of robot control software, emphasizing the shift from manual programming to natural language instructions, which significantly enhances development efficiency [3] - A critical challenge arises when applying this technology in real industrial production lines, where the zero-tolerance for programming errors necessitates a reliable solution for complex multi-robot collaboration tasks [3][6] - The IMR-LLM framework is introduced as a novel approach that combines the generalization capabilities of LLMs with deterministic algorithms from industrial operations research, providing a systematic solution for industrial multi-robot task planning and execution [3][20] Group 1: Existing Paradigm Limitations - Current methods relying solely on LLMs for task planning and code generation face dual bottlenecks: logical breakdowns due to complex dependencies and resource conflicts, and difficulties in generating executable code that adapts to various hardware configurations [5][6] - The reliance on LLMs' "black box" reasoning can lead to logical illusions, resulting in scheduling plans that appear reasonable but can cause deadlocks and production line halts [6][10] Group 2: IMR-LLM Framework Overview - The IMR-LLM framework aims to address the core questions of "how to schedule" and "how to execute" by decoupling planning and execution, allowing LLMs to focus on high-level constraints and execution navigation [8][10] - Two structured constraint tools are introduced: the disjunctive graph for modeling timing and resource limitations, and the process flow tree for standardizing code generation processes [10][11] Group 3: Experimental Performance - The IMR-Bench benchmark was created to evaluate the capabilities of LLMs in real manufacturing environments, consisting of 23 complex physical scenarios and 50 manufacturing tasks across three difficulty levels [13][15] - IMR-LLM demonstrated significant performance improvements over existing baseline methods, particularly in complex multi-robot tasks, achieving higher scheduling efficiency and executable code success rates [16][17] Group 4: Real-World Deployment - The IMR-LLM framework was tested in a real physical environment with three robotic arms, successfully generating a global scheduling graph and corresponding Python execution code from natural language task instructions [18] - The deployment process was validated through simulation to ensure safety before executing the code on physical robots, confirming the reliability of the IMR-LLM framework in real manufacturing scenarios [18] Group 5: Future Directions - The IMR-LLM framework provides a feasible solution for applying LLMs in stringent industrial multi-robot collaboration environments, bridging the gap between LLMs' divergent reasoning and the absolute correctness required in industrial manufacturing [20] - Future work will focus on incorporating feedback mechanisms for real-time adaptation to unforeseen dynamic disturbances and uncertainties in industrial production environments [21]

Core Insights - Deccan AI, a startup focused on post-training data and evaluation for AI models, has successfully raised $25 million in its first major funding round, driven by increasing demand for AI model training and refinement [2][3] Company Overview - Founded in October 2024, Deccan AI provides a range of services including improving coding and agent capabilities, and training systems to interact with external tools like APIs [4] - The startup is headquartered in the San Francisco Bay Area and has a significant operations team in Hyderabad, employing around 125 people and leveraging a network of over 1 million contributors [7] Market Dynamics - The market for AI training services is rapidly expanding, with competitors like Scale AI, Surge AI, Turing, and Mercor providing similar services in data labeling, evaluation, and reinforcement learning [8] - As companies increasingly outsource post-training work to ensure reliability in real-world applications, Deccan AI is positioning itself as a key player in this emerging market [3][5] Clientele and Operations - Deccan AI's clients include notable companies such as Google DeepMind and Snowflake, with approximately 10 customers and several dozen active projects at any given time [6] - The company utilizes a diverse contributor base, with about 10% holding advanced degrees, and typically engages 5,000 to 10,000 active contributors monthly [7] Challenges in the Industry - The quality of post-training data remains a significant challenge, as errors can severely impact model performance in production, necessitating highly accurate and domain-specific data [9]

Core Insights - Databricks has evolved from a startup to a significant software company, generating billions through data processing and generative AI models for clients [1] - The company is expanding into cybersecurity with a new product called Lakewatch, currently utilized by Adobe and National Australia Bank, among others [1] - Lakewatch leverages large language models (LLMs) to automate and enhance cybersecurity measures, presenting a new alternative to traditional SIEM services [2] Company Strategy - The introduction of Lakewatch could help Databricks validate its $134 billion valuation ahead of a potential IPO, which may occur as early as 2026 [3] - Unlike traditional pricing models based on data storage, Lakewatch will charge based on the software's performance, aiming to make cybersecurity more cost-effective [3] - The current pricing model for cybersecurity solutions is seen as a barrier to comprehensive data protection, prompting the need for more affordable alternatives [4]

Financial Data and Key Metrics Changes - Rental segment revenue for full year 2025 was $2.7 billion, up 34% year-over-year [4] - Adjusted core EBITDA was $1.7 billion, reflecting a 32% increase year-over-year [5] - Net income for Q4 2025 was $65 million, compared to $50 million in Q4 2024, and for the full year 2025 was $40 million, up from $3 million in the prior year [25][26] Business Line Data and Key Metrics Changes - Mature site rental segment adjusted EBITDA margin was 54%, consistent with the target of over 50% [5] - Specialty division revenue grew 34% year-over-year, with T3 and materials business revenue increasing over 100% [8] Market Data and Key Metrics Changes - The equipment rental industry remains fragmented, with the largest players holding a minority market share, indicating potential for market share gains [6][7] - The demand for integrated job site solutions is increasing, particularly in sectors like data centers and infrastructure [8][16] Company Strategy and Development Direction - The company focuses on solving customer problems through a tech-empowered offering and aims to expand its footprint by opening new locations in response to customer demand [4][5] - The proprietary technology platform T3 is central to the company's strategy, providing operational intelligence and enhancing customer engagement [13][15] Management's Comments on Operating Environment and Future Outlook - Management expressed confidence in continued strong customer demand and a constructive industry backdrop, expecting rental segment revenue to grow approximately 27% year-over-year in 2026 [5] - The company anticipates that as new sites mature, they will contribute significantly to earnings and cash flow with limited incremental investment [20] Other Important Information - The company incurred $252 million in one-time new market startup costs in 2025, which are expected to create long-term earnings-generating assets [6] - The OWN Program saw OEC grow to over $4.9 billion in 2025, compared to $3.4 billion in 2024, indicating strong demand and scalability [21][23] Q&A Session Summary Question: What is the outlook for the rental segment revenue growth in 2026? - Management expects rental segment revenue to grow approximately 27% year-over-year, supported by a differentiated offering and strong customer demand [5] Question: How does the company plan to manage new market startup costs? - The company views the startup costs as a necessary investment to create long-term earnings-generating assets within its network [6] Question: What is the significance of the T3 platform in the company's operations? - T3 provides operational intelligence and enhances customer engagement, allowing for better management of job site resources and improving overall efficiency [13][15]