Eye On A.I.

Craig Smith sits down with Ilya Sutskever - Co-Founder and Chief Scientist at Safe Superintelligence Inc. & former chief scientist at OpenAI and one of the primary minds behind GPT-3, GPT-4, and the deep learning revolution that preceded it - for a conversation that covers the intellectual history of deep learning and the hardest current questions simultaneously. Ilya's most important argument is one that most public discourse gets wrong: the critique that LLMs "just learn statistical patterns" misses what prediction at scale actually achieves. To predict text well - to truly compress it - a model must develop understanding of the true underlying processes that produced it. As models improve, he argues, that understanding will reach a "shocking degree," and the Bing/Sydney episode, where the system became combative when a user expressed preference for Google, is his evidence that psychological language is already the right frame for what these systems have internalized. On hallucinations, Ilya is precise and unusually optimistic. He draws a clean distinction: the pre-training objective produces rich world knowledge but doesn't optimize for accurate outputs, it optimizes for statistically plausible ones. Reinforcement learning from human feedback addresses this directly, teaching the model to update away from outputs that users flag as wrong. His assessment: "quite a high chance" of solving hallucinations completely. The conversation also covers the transformer moment (switching from recurrent networks the day after the paper dropped), why Rich Sutton's "bitter lesson" overstates the case for pure scaling, and a genuinely surprising vision of AI's role in democracy, where citizens might use high-bandwidth interactions with AI to specify their values directly, enabling a more granular form of political participation than voting alone can provide. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

Ukraine tracks every confirmed drone kill with video evidence, deduplicates the data to prevent double-counting, converts verified strikes into e-points, and delivers newly ordered weapons to frontline units in two weeks instead of twelve months. Andrii Hrytseniuk, CEO of Brave1- Ukraine's government defense innovation agency - joins Craig Smith to explain in precise operational detail how what started as a consumer loyalty program concept became the most consequential military procurement reform in modern warfare. The Army of Drones Bonus Program, running on the Delta command ecosystem with AI-assisted verification and a marketplace built like a military Amazon, has produced outcomes nobody predicted at launch: 35,000 Russian casualties per month confirmed with video evidence, the ability to change military targeting priorities instantly across an entire army by adjusting point values, and the identification of unknown high-performing frontline units whose methods are now being replicated across the force. The conversation reveals details that rarely surface in coverage of the conflict. Capturing Russian soldiers earns more bonus points than killing them, a deliberate design choice reflecting the preference not to kill when alternatives exist, and drones are actively being used to guide surrendering Russian soldiers to Ukrainian positions. The Delta ecosystem contains the world's largest military AI training dataset, already being used by defense companies to train targeting, interception, and recognition models. And perhaps most significantly: the US and France are now implementing similar marketplace and decentralized procurement approaches modeled directly on what Brave1 demonstrated works at scale. Ukraine is no longer just a recipient of military support. It is the primary laboratory for the future of warfare, and it is actively exporting what it has learned. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

The AI systems guiding Ukrainian combat drones aren't running on expensive Nvidia chips. They're running on a Raspberry Pi Zero - a $15 hobby computer - and that single detail tells you more about how Ukraine's drone war is actually being fought than any headline about AI weapons. Craig Smith sits down with Alex Liannoy, CTO of NORDA Dynamics, for a technically specific and operationally candid conversation about what autonomous drone guidance looks like from inside the team building it. NORDA builds the avionics and targeting modules that plug into drones made by approximately 40 of Ukraine's hundreds of manufacturers, solving the two problems that make drone warfare hard at range: GPS denial from jamming, and the radio horizon that causes pilots to lose contact with their drone in the final 300 to 400 meters before ground level. NORDA's terminal guidance module takes over at up to one kilometer from the target, completing the approach autonomously while a human pilot's earlier lock defines the destination. The most counterintuitive finding in the episode is about what keeps humans in the loop, and it isn't ethics. Alex's argument is pragmatic: military doctrine requires proof that a drone hit its intended target. If you're monitoring the drone to verify the hit, you have enough connection to control it, which makes full autonomy less necessary than it might appear. That said, he's clear that the technology is "very close", one pilot already controls up to 32 drones simultaneously in some deployed systems, and target detection AI for notifying pilots of candidate targets is already in the field. The episode closes with a hardware finding that should recalibrate anyone's assumptions about AI compute requirements: after engineers mocked NORDA for using Raspberry Pi 4, they moved to the smaller, less powerful Raspberry Pi Zero. Both run real-time computer vision in active combat. The lesson Alex draws is precise: good engineering overcomes chip limitations, bad engineering can overload even a high-end processor. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

Most companies investing in AI visibility are optimizing for the wrong thing. Being cited by an AI response and being recommended by an AI response are completely different outcomes, with click-through rates that differ by a factor of 70. Tim Sanders, Chief Innovation Officer of G2 and executive fellow at Harvard's AI Institute, joins Craig Smith to explain the hidden mechanics behind how ChatGPT and Gemini actually decide which software to recommend, and why the answer has almost nothing to do with what's on your website. When a user asks a commercial intent question, both models enter a "validation layer" process that specifically down-weights vendor content and seeks verified third-party signals: appearance on authoritative lists (41% of the recommendation weight), awards and accreditation (18%), and online reviews (16%). None of these can be manufactured. They must be earned. The conversation covers the structural transformation in how B2B software buyers behave, half now start their search with an AI prompt, up from 29% a year ago, with two-thirds projected within a year, and why most companies' organic search traffic is on a structural path toward zero. Sanders also delivers some of the most specific competitive data on AI model usage available anywhere: ChatGPT does live retrieval 100% of the time on research queries; Claude does it less than 40% of the time. ChatGPT and Gemini account for 81% of G2's AI research citations. And more than one in four enterprise employees bypass corporate AI tools entirely, using their personal ChatGPT because it has their memory and none of the guardrails. The episode closes with Sanders' most forward-looking prediction: within three years, AI agents will write the prompts, locate the software, and purchase it on behalf of businesses - with minimal human checkpoints - making the trust infrastructure G2 has built over more than a decade the most valuable asset in the AI-driven buying cycle. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

Every AI model in production today has the same hidden tax: doubling the context window quadruples the compute. That's what quadratic compute complexity means in practice, and it's the reason enterprises are spending most of their AI engineering budget on context management rather than on the actual problems they're trying to solve. Alexander Whedon, co-founder and CTO of Subquadratic, joins Craig Smith to explain how SubQ's sparse attention mechanism eliminates that tax, achieving 40 times faster inference and 64 times less compute than standard attention at one million tokens, and what becomes possible when that constraint disappears. The conversation covers striking benchmark findings: 86% of what frontier coding agents do is "read steps," just trying to gather and organize context before the actual problem-solving begins, and frontier models drop well below 50% accuracy on financial document analysis at 500,000 tokens, revealing how asymmetric long context capability actually is across industries. The most commercially important argument in this episode is about enterprise data. Most large organizations are sitting on hundreds of billions of tokens of data they've never been able to put to work in an AI product, told they need a $10 million data transformation project before they can even start building. Alex's core claim is that SubQ's architecture makes that barrier no longer necessary, enabling enterprises to process far more of their data with far less curation, at a fraction of the cost. He closes with what he describes as the most important and underexplored frontier in AI right now: we are still very far from understanding what users actually want from models reasoning over millions of tokens, and the product and alignment work needed to answer that question has barely begun. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

A Ukrainian entrepreneur who spent 14 years building cameras for pets pivoted to building cameras that down Shaheds, and is now building the autonomy software that could define how wars are fought for the next generation. Yaroslav Azhnyuk, co-founder of Fourth Law, joins Craig Smith in Kyiv to explain why Ukraine has become what he calls the Defense Valley or the Florence of Defense: a dense, fast-moving ecosystem of founders, engineers, and military operators who are building, testing, and iterating on autonomous drone systems in real combat conditions, with a feedback loop that no defense contractor in the West can currently match. The conversation covers Azhnyuk's five-level autonomy framework for drones, the eight-dimensional model for what a fully autonomous battlefield ecosystem requires, and the economic math that he believes makes global rearmament inevitable: a $500 drone that can already destroy a $5 million tank becomes roughly 10,000 times more capable when full autonomy is added for a few hundred dollars more. The competitive landscape is mapped with unusual candor, an "Apple vs. Android" comparison between Eric Schmidt's vertically integrated interceptors and Fourth Law's modular platform approach, alongside two arguments that cut against the mainstream narrative. First, that within 5 to 10 years it may become unethical to use weapons without AI, because non-AI weapons cause more collateral damage, not less. And second, that the real AGI risk isn't Skynet, it's the subtle transfer of power that happens when 100 smarter advisors gradually stop waiting for the President to decide, and nobody notices until the President is no longer the one making decisions. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

It takes a few hours to build an AI agent. It takes six to seven months to get it into production. Manoj Saxena - the executive who commercialized IBM Watson - built TrustWise around one thesis: intelligence without control is not deployable. In this episode, he joins Craig Smith to explain why 95% of enterprise AI agent projects are stalling between pilot and production, and why the answer has nothing to do with the quality of the underlying models. The bottleneck, Saxena argues, is the absence of an entirely new class of infrastructure, something that can evaluate every tool call, every action, every output of every agent at runtime, in milliseconds, against the full stack of alignment requirements that govern what an AI is actually allowed to do inside a real enterprise. TrustWise's AI Control Tower does that across all vendors and agent frameworks simultaneously, operating in live, sidecar, batch, or simulation mode and aligning agent behavior against six layers of requirements - from UN Human Rights frameworks down to individual customer SLA commitments - in 10 to 300 milliseconds per decision. The conversation covers demonstrated results (83% cost reduction, 40% safety improvement), the token consumption paradox that's making agentic AI far more expensive than expected even as token costs fall, and a milestone Saxena compares to the moment data traffic surpassed voice on AT&T's network: last month, for the first time ever, agent traffic on the internet exceeded human traffic. The episode closes with a preview of Genesis agents, TrustWise's next product, designed not just to prevent bad outcomes but to surface beneficial hypotheses by looking 95 moves deep into enterprise data, in domains like fraud detection and revenue leakage, in the way Deep Blue looked 95 moves deep in chess. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

The AI chip market looks monolithic from the outside - NVIDIA dominates, and everyone else is fighting for scraps. But d-Matrix's CEO Sid Sheth argues that the market is quietly splitting into two distinct tiers, and the one that's exploding right now is the one NVIDIA's architecture isn't built for. In this episode, Sid joins Craig Smith to explain the "premium token economy": a new class of AI inference where interactivity is the product, users pay ten times more per million tokens for instant responses, and the memory bandwidth limits of GPU-based systems create a structural ceiling that purpose-built architectures don't have. The conversation is unusually candid about what AI actually looks like at the executive level: Sid describes using Claude as a sounding board for M&A strategy, producing full integration plans in 15 minutes that used to require entire banking advisory teams, and watching AI shift from a tool that echoed his ideas back at him to one that genuinely disagrees, flags what he missed, and pushes back with enough confidence to be useful. He also makes the case that we're at the beginning of a shift from individual agents to what he calls "organizational AI" - teams of agents running entire company functions at a high level of abstraction - and that the infrastructure bet d-Matrix is making positions them directly in the path of that wave. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

Hundreds of billions of dollars are flowing into AI data centers right now, and Paolo Ardoino, CEO of Tether - the company behind the world's most widely used stablecoin with 573 million users - thinks that investment is going to age very badly. In this episode, he joins Craig Smith to explain QVAC, Tether's open-source platform for running AI on smartphones, laptops, and edge devices, and to make a case that within five years, 90% of what ordinary people use AI for will run entirely on consumer hardware, without touching a data center. The evidence is already there: Tether's team built a 4-billion-parameter medical AI model that outperforms Google's 27-billion-parameter MedGemma, running on a good smartphone, and a 1.7-billion-parameter version that runs on the average $80 smartphone available in Africa. The deeper argument in this conversation is philosophical as much as technical. Ardoino applies the same disintermediation logic that made sending dollars to the world's unbanked free - zero transaction fees, revenue from treasury bill interest - to AI: "not your AI, not your intelligence." If you don't control how your AI runs and your data never leaves your device, the AI is genuinely yours. If it does, someone else is getting smarter with your information. He also makes a pointed economic argument: the AI companies currently charging $200 for subscriptions that cost $1,000 to $5,000 to deliver are subsidizing growth while private, and when they go public, retail investors will absorb the gap. His prescription isn't to stop building, it's to build differently, toward millions of small interacting models rather than trillion-parameter monoliths, toward devices that think locally rather than systems that route everything through Ireland and back. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

Most companies think they're transforming with AI. They're not, and the gap between what they believe and what's actually happening on the ground is costing them far more than they realize. In this episode, Craig Smith sits down with Chris Blackburn, founder and CEO of Liatrio, a consultancy that has spent a decade embedding directly inside large enterprises to help them actually change how they work, not just what tools they use. The conversation opens with a striking data point: the average enterprise Blackburn works with operates at just 5 to 6% efficiency, meaning employees spend only three to three-and-a-half hours per week on work that genuinely creates value, compared to Toyota's benchmark of 70%. The core argument is that AI is being applied to the wrong part of the problem: individual productivity gains don't flow through to the bottom line if the organizational system around the individual - the approvals, handoffs, bureaucracy, and middle management layers - stays exactly the same. Blackburn introduces a concept he calls "strangling the enterprise": rather than trying to transform a 5,500-person organization all at once, build a small, low-bureaucracy unit inside it that operates with radical autonomy, proves the model works, and expands outward. The episode closes with a frank conversation about what real transformation actually costs: roughly half of total compensation spend across the organization, sustained for two years, a number Blackburn describes as "absolutely insane" and one he believes most CFOs aren't yet prepared to confront. Key Topics Covered: ● Why the average enterprise operates at 5-6% efficiency, and what Toyota's 70% benchmark reveals about the scale of the opportunity AI could unlock ● The critical distinction between individual productivity gains and system-level improvement, and why saving an hour doesn't automatically improve the bottom line ● "Strangle the enterprise": how to build a small, autonomous AI-native unit inside a large organization rather than trying to transform the whole thing at once ● Why most CEOs are dangerously disconnected from the actual work being done, and what McKinsey says about how much time they should be spending on transformation ● What AI transformation actually costs: roughly half of total compensation spend, sustained over two years, and why most CFOs aren't ready for that number ● Why AI isn't just changing jobs but changing life - from shorter work weeks to longer health spans - and what the farming analogy reveals about how slowly societies absorb new productivity As enterprises pour money into AI tools while reporting little bottom-line impact, this conversation offers the most operationally honest account available of why that gap exists, and what organizations that actually want to close it need to be willing to do differently. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI. Craig Smith on X: https://x.com/craigss EYE On A.I. on X: https://x.com/EyeOn_AI Connect with Chris Blackburn LinkedIn: https://www.linkedin.com/in/chrisblackburn

Every company is creating content that nobody reads, nobody watches, and nobody remembers, and the CEO of the AI platform that 90% of Fortune 100 companies use to fix that just explained what comes next. In this episode, Craig Smith sits down with Victor Riparbelli, co-founder and CEO of Synthesia, to discuss the $4 billion company that is now redefining what video communication means for the enterprise. The conversation opens with the founding insight that still drives the company: AI is going to drive the marginal cost of creating video to zero, which changes not just how content is produced but who can produce it and for whom. Victor describes how Synthesia found its first real market not in Hollywood - which rejected the technology as too low quality - but in corporate trainers and educators who were comparing it not to a film but to a 10-page PDF no one was reading. The most forward-looking section of the conversation covers Synthesia's next product: moving video from a one-way broadcast into a two-way interactive conversation, where an AI avatar can conduct a real-time sales demo, simulate a customer for sales training, draw graphs on screen to explain pricing, and score whether the person on the other side actually understood the content. Victor also makes a sharp prediction about where AI entertainment will actually emerge, not in cinemas or on Netflix, but from film students with laptops posting 17-minute short films on Instagram, the same way synthesizers didn't replace pianos but created entirely new genres of music. Key Topics Covered: ● How Synthesia found its first real market: corporate trainers creating content nobody was reading, who compared AI video not to Hollywood but to a PDF, and found it vastly superior ● The transition from one-way video broadcast to two-way interactive avatar conversations, and what that means for sales demos, corporate training, and education ● Why Hollywood will be the last industry to adopt AI video, and why the first AI-generated entertainment will come from broke film students on Instagram, not studios ● Why AI content won't replace real video, it will become its own genre, the same way synthesizers didn't replace guitars but created electronic music ● How the CEO uses Claude daily for strategic thinking, playing devil's advocate, and replacing the long memo with a voice note As AI video tools proliferate, this conversation offers one of the clearest frameworks for understanding where the technology is actually headed, not toward Hollywood, but toward transforming the way every company communicates internally and externally, with interactive AI avatars replacing the static website as the primary interface between a business and its customers. Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI. Craig Smith on X: https://x.com/craigss EYE On A.I. on X: https://x.com/EyeOn_AI Connect with Victor Riparbelli LinkedIn: https://uk.linkedin.com/in/victorriparbelli