LLM vs Fact Models in AI Applications

I keep hearing “ChatGPT-5 will replace solver engines.” That’s not how this works. Let’s break it down: 🔹 LLMs (like ChatGPT) LLMs are probabilistic pattern generators. They predict the next most likely word or token, which makes them great at writing text, drafting code, and summarizing knowledge. But they do not guarantee: • Feasibility → whether a solution satisfies all constraints. • Optimality → whether no better solution exists. • Correctness → whether the answer is even valid in a mathematical sense. Every output is, at best, a plausible guess. 🔹 Optimization solvers (Gurobi, CPLEX, CBC, etc.) Solvers are deterministic engines. They take a mathematical model (variables, constraints, and an objective) and: • Explore massive search spaces (millions or billions of possibilities). • Use decades of algorithmic advances (branch-and-bound, cutting planes, decomposition). • Prove feasibility and, often, prove optimality. This difference is crucial. ⸻ ✅ Example 1: Truck routing with time windows 👉 LLM: can generate a MILP formulation or pseudocode. 👉 Solver: systematically searches the combinatorial explosion of routes, ensuring trucks don’t violate capacity or timing rules, and finds the least-cost solution. ✅ Example 2: Portfolio optimization 👉 LLM: can describe the model and constraints in plain English. 👉 Solver: ensures budgets are not exceeded, risk constraints are respected, and returns the provably best allocation of capital. ⸻ ✅ The key distinction: LLMs are model assistants. Solvers are solution engines. LLMs help translate messy business problems into math. Solvers deliver mathematically rigorous answers. The future isn’t replacement, but rather synergy. Use AI to frame the problem, then let optimization engines do what they’re built for: find the best decision under constraints.

Beyond GenAI: Why the Future Is LLMs + Large Quantitative Models Recent industry signals (including Salesforce's open acknowledgment of reduced confidence in GenAI for core decisioning) aren’t a rejection of AI. They reflect a growing realization: language intelligence alone is not enough for enterprise decisions. What’s being questioned isn’t the value of LLMs, but the assumption that stochastic generation can replace predictive reasoning. LLMs are exceptional at synthesis, explanation, and interaction. They transform how insights are consumed. But they are not predictive systems. They don’t model uncertainty explicitly, produce calibrated probabilities, or learn directly from outcomes. That makes them powerful assistants and risky standalone decision engines. At the other extreme, deterministic automation (RPA) is reliable but brittle. It encodes yesterday’s assumptions and struggles under change. The real opportunity lies in the middle: Large Quantitative Models (LQMs). LQMs explicitly quantify uncertainty, reason across hierarchy (deal → account → segment → company), operate across time (current, next, future horizons), and continuously learn from outcomes. They provide the mathematical backbone needed for trustworthy decision-making. The future isn’t LLMs or LQMs. It’s LLMs + LQMs, with a clear division of responsibility: LQMs predict outcomes, quantify risk, measure error, and enforce learned guardrails LLMs explain predictions, surface tradeoffs, and translate quantitative outputs into action In short: LQMs decide. LLMs explain and assist. This is why platforms built on predictive foundations are better positioned for GenAI. When guardrails, accountability, and learning loops already exist, language intelligence becomes additive — not risky. As the industry moves past GenAI experimentation, the platforms that win will: Reason forward, not summarize backward Measure their own error and improve over time Earn trust through accountability, not fluency LLMs make AI accessible. LQMs make it reliable. Together, they make AI deployable at scale. https://lnkd.in/gXktR5nn

Large Language Models (LLMs) and Large Reasoning Models (LRMs) represent two evolving paradigms in AI. Both are powerful, but they differ in architecture, training objectives, and how they perform in complex “agentic” workflows. In agentic AI systems - where an AI agent plans actions, uses tools, or delegates tasks - choosing the right model for the orchestration layer is critical. This post dives deep into how LLMs (e.g. GPT-4, Anthropic Claude, Mistral) compare with LRMs (e.g. smaller instruction-tuned models, Mixture-of-Expert architectures) and why their differences matter for building intelligent agents. LLMs predict, while LRMs reason. An LLM responds immediately based on learned patterns, whereas an LRM pauses to think - it sketches a plan, weighs options, double-checks calculations (often via tools or sandboxed tests), and verifies intermediate results before answering. This fundamental difference means LRMs can tackle complex logical problems more systematically, whereas LLMs may sometimes hallucinate or make reasoning errors if a problem requires steps that weren’t implicit in their training. On the flip side, LLMs tend to be faster and more creative for straightforward tasks, since they don’t spend time “overthinking” when it’s not necessary.

AI researchers’ obsession with benchmarks that don’t measure real-world performance is destroying the field’s credibility. As model sizes rise, and costs spiral up, neither functionality nor reliability significantly improves. Small language models lag LLMs in the benchmarks, but run them head-to-head in a product, and users can’t tell the difference. I have seen this with my clients for almost two years. LLMs aren’t worth the price tag. There’s a huge contradiction. The bigger the model, the better it performs on AI benchmarks, but that hasn’t translated into similar improvements on real-world tasks. SLMs and LLMs are near parity. Hallucinations have only marginally diminished as LLMs have spent the last three years making massive leaps in scale. PhD-level benchmark performance still fails most common sense reasoning tests. Larger models can’t explain how they arrived at their outputs reliably either. That dramatically increases the cost of tracking the root causes of errors and resolving them. With both SLMs and LLMs, we must put the same information guardrails in place to drive functional gains. Without a knowledge graph, neither LLMs nor SLMs do anything reliably enough to put in front of customers, so why not use the lower-cost option? That’s what an increasing number of AI vendors are choosing. Salesforce and Microsoft were both early to switch from LLMs to SLMs. Training and inference costs plummet, making more use cases feasible. Business leaders must reevaluate their AI action plans to ensure their budgets deliver the highest value possible. SLMs drop the per-initiative cost, so firms can deliver more initiatives per year, creating higher ROI. SLMs don’t require the most expensive AI talent, and the available talent pool is much larger. The barriers to ramping up an AI team are much lower, and again, compensation costs drop as well. AI is an information product that requires new information architecture and knowledge management systems. Businesses must factor the costs and level of effort into their AI action plans. However, investments into knowledge management systems have some of the highest long-term ROIs. Each part of the knowledge graph can be monetized multiple times and returns compound with each new model. 🔔 Keeping up with the leading edge of AI and gaining first-principles expertise is simple. Turn on notifications by ringing the bell in my profile. 🔔

Everyone assumes LLMs are the future. NVIDIA & Georgia Tech just made the case for the opposite. After digging into their new, provocative paper: 𝑆𝑚𝑎𝑙𝑙 𝐿𝑎𝑛𝑔𝑢𝑎𝑔𝑒 𝑀𝑜𝑑𝑒𝑙𝑠 𝑎𝑟𝑒 𝑡ℎ𝑒 𝐹𝑢𝑡𝑢𝑟𝑒 𝑜𝑓 𝐴𝑔𝑒𝑛𝑡𝑖𝑐 𝐴𝐼, one message is clear: We do not always need massive LLMs to build effective AI agents. The paper makes three bold claims: 𝟏. 𝐏𝐨𝐰𝐞𝐫𝐟𝐮𝐥 𝐞𝐧𝐨𝐮𝐠𝐡: SLMs can handle tool use, instruction following, code generation, and reasoning, core tasks for AI agents. 𝟐. 𝐎𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐟𝐢𝐭: Agents mostly need decision-making (e.g., “which tool to call”), not essays or poetry. SLMs are optimized for such focused tasks. 𝟑. 𝐂𝐡𝐞𝐚𝐩𝐞𝐫: A 7B model costs 10–30x less than a 70B model, consumes less energy, and can run locally. So how exactly do they define a Small Language Model (SLM)?  → An SLM is compact enough to run on consumer devices while delivering low-latency responses to agentic requests.  → An LLM is simply a model that is not an SLM. Supporting arguments from the paper: → 𝐂𝐚𝐩𝐚𝐛𝐢𝐥𝐢𝐭𝐲: Modern SLMs rival older LLMs in reasoning, instruction following, and tool use, and can be boosted further with verifier feedback or tool augmentation. → 𝐄𝐜𝐨𝐧𝐨𝐦𝐢𝐜𝐬: They are dramatically cheaper to run, fine-tune, and deploy, fitting naturally into modular, “Lego-like” architectures. → 𝐅𝐥𝐞𝐱𝐢𝐛𝐢𝐥𝐢𝐭𝐲: Lower costs make experimentation easier and broaden participation, reducing bias and encouraging innovation. → 𝐏𝐫𝐚𝐜𝐭𝐢𝐜𝐚𝐥 𝐟𝐢𝐭: Agents only need narrow functionality like tool calls and structured outputs. LLMs’ broad conversational skills often go unused. → 𝐀𝐥𝐢𝐠𝐧𝐦𝐞𝐧𝐭: SLMs can be tuned for consistent formats (like JSON), reducing hallucinations and errors. → 𝐇𝐲𝐛𝐫𝐢𝐝 𝐚𝐩𝐩𝐫𝐨𝐚𝐜𝐡: LLMs are best for planning complex workflows; SLMs excel at execution. → 𝐂𝐨𝐧𝐭𝐢𝐧𝐮𝐨𝐮𝐬 𝐢𝐦𝐩𝐫𝐨𝐯𝐞𝐦𝐞𝐧𝐭: Every agent interaction generates training data, allowing SLMs to steadily replace LLM reliance over time. Of course, skeptics argue that LLMs will always outperform due to scaling laws, economies of scale, & industry inertia. But the authors make a strong case that SLMs are cheaper, faster, specialized, and sustainable. And the best part? They openly invite critique and collaboration to accelerate the shift.  

Back to the Future: from LLM Hype to Agentic Reality With the progress of LLMs and the pivot to agentic AI, we’re finally acknowledging an important truth: LLMs are powerful but not reliable decision-makers. They excel at natural language interaction and orchestrating workflows but when it comes to reliable, auditable and repeatable decisions, we’re back to the fundamentals: machine learning models. It’s a return to balance: - LLMs for interaction and adaptability. - ML models for robust predictions, decision-making and optimization. In practice, this means: LLMs as agents that call models. ML models as the decision engines powering trustworthy outcomes. A new generation of hybrid systems where interpretability and performance reliability can be engineered with rigor. The era of “just use LLM” is closing. The era of building with LLMs grounded in machine learning foundations is here to stay.