Oct 30, 2024 · 50m · big-technology
The Next Gen AI Models: Reliable, Consistent, Trustworthy — With Cohere CEO Aidan Gomez
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Cohere CEO and transformer co-author Aidan Gomez joins Alex Kantrowitz to demystify generative AI hype, arguing that real-world value lies in building reliable, cost-effective models that augment human workers and automate enterprise workflows.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 36.6% of the talking time here. How this is scored →
Alex as informed peer 5.4 Guest teaching 3.9 Guest disagreement 2.2 Alex pushing back 3.7
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 37:36 Aidan forcefully refutes the trivialization of back-office enterprise AI use cases
Aidan rejects Alex's devil's-advocate critique that automated drafting and search are low-value conveniences, firmly countering that supply chain risk is measured in trillions and professional billable hours make mundane automation immensely lucrative.
Hardest push from Alex ▶ 36:41 Alex channels AI critics to challenge multi-billion dollar capex for routine administrative tasksAlex aggressively adopts the role of an AI skeptic, questioning why billions in capital are pouring into software that merely automates job descriptions and standard vendor searches that skilled humans already do well.
Biggest teaching moment ▶ 23:05 Aidan explains model interpolation mechanics and dispels intelligence explosion mythsAidan provides deep technical instruction on why models cannot magically discover concepts beyond their training distribution, explaining that self-improvement through synthetic data reaches a hard plateau rather than causing runaway superintelligence.
Alex holds their own ▶ 9:25 Alex demonstrates cluster knowledge contrasting Llama 3 with Elon Musk's supercomputerAlex brings granular industry figures to the table, contrasting Meta's 16,000 GPU training cluster with xAI's 100,000 GPU Memphis supercluster to drill down into the practical returns of massive compute scaling.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| The Sustainability and Upfront Costs of AI Scaling | 6 | 3 | 1 | 4 | Alex opens with concrete financial reporting regarding OpenAI's $6.6 billion raise against a $5 billion annual burn rate, questioning the macro-economic sustainability of escalating compute and energy costs. Aidan calmly contextualizes this as massive upfront capital expenditure that transitions from initial proofs-of-concept into scalable enterprise ROI. | |
| Debunking God-Like AI Models in Favor of Reliability | 5 | 4 | 4 | 5 | Aidan immediately dismisses the hype term 'godlike AI' and urges setting aside expectations of omnipotent AGI. Alex maintains pressure by asking pointedly what tangible, measurable capabilities will actually emerge from the upcoming generation of heavily funded models. | |
| High-Stakes Applications and Building Accuracy in Healthcare AI | 6 | 3 | 2 | 4 | Alex offers firsthand anecdotal reporting from a clinician consulting AI for medical treatment, contrasting 60% versus 90% threshold reliability and probing whether steady incremental progress might puncture the industry's hype bubble. Aidan emphasizes that immense economic value exists even if underlying model technology frozen today. | |
| Compute Cluster Scaling and Diminishing Returns on Hardware | 7 | 4 | 2 | 3 | Alex demonstrates deep domain knowledge by citing specific hardware metrics, contrasting Meta's Llama 3 training cluster of 16,000 GPUs with xAI's 100,000 GPU Memphis supercluster. Aidan details the physics and economics of hardware scaling, explaining that while scale yields smarter models, returns saturate and cost-efficiency dictates right-sizing. | |
| Defining Artificial General Intelligence Versus Superintelligence Expectations | 5 | 3 | 2 | 4 | Alex delineates between task-based artificial general intelligence and theoretical superintelligence, challenging the colloquial conflation of the two terms. Aidan concurs that achieving human parity on discrete tasks is a grounded and achievable engineering objective. | |
| The Future of Labor, Workforce Automation, and Meaning | 5 | 4 | 3 | 6 | When Aidan asserts that technological adoption will not cause mass unemployment due to infinite economic demand, Alex directly challenges the thesis, questioning why humans would continue working if knowledge tasks can be automated. Aidan reframes labor as a source of purpose and intellectual fulfillment. | |
| Reasoning Models and Explainable Chains of Thought | 5 | 4 | 3 | 5 | Alex questions why major AI leaders signed open letters comparing AI safety risks to nuclear war and pandemics. Aidan distances himself from existential doomsday narratives, noting he intentionally refused to sign the statement and explaining how chain-of-thought reasoning enhances inspectability and control. | |
| Emergent Behaviors and the Myth of Intelligence Explosion | 4 | 7 | 4 | 2 | Aidan demystifies the popular notion of emergent behaviors and runaway intelligence explosions, explaining that LLMs function as skill interpolators across existing domains. He educates the host on how self-improving synthetic data loops invariably hit plateaus rather than accelerating indefinitely. | |
| Human Data Bottlenecks and Strategic Use of Synthetic Data | 6 | 5 | 2 | 4 | Alex recalls Aidan's past comments regarding the human data bottleneck moving from general crowdsourcing to hiring subject-matter PhDs. Aidan clarifies where synthetic data works (verifiable domains like code and mathematics) versus where it fails (subjective fields like philosophy and social sciences). | |
| The Embodiment Debate: Can Observation Alone Achieve Intelligence? | 5 | 5 | 3 | 2 | Alex raises academic debates over whether passive language modeling requires physical robotic embodiment to achieve real-world understanding. Aidan articulates a clear contrarian stance, arguing that multimodal observation across the internet is sufficient to achieve human-level intelligence. | |
| Podcast Intermission and Mid-Roll Sponsor Message | 3 | 2 | 0 | 1 | A mid-roll sponsor break transitions into a brief retrospective where Alex asks Aidan about co-authoring the seminal 'Attention Is All You Need' paper and whether Google effectively commercialized the transformer architecture. | |
| Realizing Enterprise Return on Investment with Oracle Integration | 5 | 4 | 1 | 4 | Alex pushes for concrete, verifiable examples of enterprise return on investment rather than abstract promises. Aidan details Cohere's partnership with Oracle, where generative AI powers over 50 production enterprise applications across HR, legal, and supply chain management. | |
| The Economic Scale of Mundane Back-Office Task Automation | 7 | 4 | 6 | 7 | Alex plays devil's advocate, challenging whether minor productivity gains in job descriptions or contract review justify billions in capital investment. Aidan forcefully counters that supply chain risks represent trillions of dollars and expensive professional billable hours make mundane automation tremendously valuable. | |
| Consulting Growth and Enterprise AI Diffusion with Accenture | 7 | 3 | 1 | 2 | Alex introduces reporting from Reuters showing Accenture's generative AI bookings grew 50% quarter-over-quarter despite broader consulting slowdowns. Aidan explains how systems integrators act as key distribution channels to embed models into legacy corporate infrastructure. | |
| Workforce Augmentation and the Reality of Employee Replacement | 6 | 3 | 2 | 5 | Alex directly presses Aidan on whether any enterprise client is utilizing Cohere's software to eliminate full-time headcount. Aidan firmly maintains that current implementations focus exclusively on worker augmentation and task acceleration rather than direct job displacement. | |
| Cloud Providers, On-Premise Security, and Retrieval-Augmented Generation | 6 | 4 | 1 | 3 | Alex cites CNBC financial breakdowns showing Anthropic generating the majority of its revenue via cloud distribution partners like AWS. Aidan explains Cohere's dual architecture, balancing hyperscaler cloud partnerships with on-premise deployments and specialized Retrieval-Augmented Generation (RAG). | |
| Two-Year and Five-Year Predictions for Autonomous AI Agents | 4 | 4 | 1 | 2 | Alex concludes by asking for concrete multi-year timelines for autonomous agent capabilities. Aidan outlines a progression from near-term daily assistive collaborators to deeply integrated, high-competency autonomous agents over a five-year horizon. |
Statements from this episode (27)
Gomez: AI Training Costs Are Small Relative to Long-Term Value
“I certainly understand the urge for people to see the numbers being spent on training and be concerned that it's not going to recoup in value, but I think that Those numbers are actually small relative to the long-term value that the technology will deliver.”
Gomez: 2023 Was AI Proof of Concept; 2024 Is Scale Production
“So last year was very much the year of the proof of concept. People were getting familiar with the technology. It was their first time working with it. And so there were a lot of small tests and experiments, but this year is very much one of going to productio…”
Gomez: AI Industry Is Building Useful Tools, Not an AGI God
“I don't know about God. Like, I don't think I'd ever use that term to describe what's coming. I think we're going to have some really powerful and useful tools. Emerge. I think that's what's coming. The idea that we're building AGI or something that's just gon…”
Gomez: Next-Gen AI Models Will Break Far Less Frequently
“I expect them to be far more robust, reliable and I expect that they'll just be more capable. There's lots of things today that models break in unsupp, in surprising ways, and I think that's going to start to become rarer and rarer. We're going to be able to p…”
Gomez: AI Progress Will Be a Continuous Curve, Not a Step Change
“I don't see a step change coming, but I see a steady continuous course towards very high accuracy, very high reliability AI.”
Gomez: Frozen AI Tech Would Still Yield Incredible Economic Value
“Even if, you know, like just as a hypothetical even if the technology froze and what we have today is all we get, there's so much good to be done. There is so much work to go do. To implement this technology across the economy really boost productivity, drive …”
Gomez: Massive AI Models Are Useless If Too Expensive To Deploy
“My personal perspective is that, you know, building a massive model, it's not actually useful for the world if it's too big to be consumed, if it's too expensive to actually deploy.”
Gomez: AI Scaling Gains Will Suffer Diminishing Returns
“I don't think within any achievable scaling up for humanity that will reach that tipping point. It just saturates the gains become much, much smaller. And so you're much less willing to want to pay double the price for a minute difference. But it is pretty con…”
Gomez: AGI Matching Human Performance Is Achievable and Reasonable
“I mean, with that definition of AGI, I think it's both Achievable and a fairly reasonable target.”
Gomez: AI will never cause mass unemployment of humans
“Well, because I think that we will never see mass unemployment of humans. I think that this technology is going to unlock more opportunities. It will let us do more as opposed to Scaling back what we do.”
Gomez: AI Productivity Expands Total Output in a Supply-Constrained World
“Humanity is very supply side constrained, not demand side. We always, we want more. We want better. We want to be healthier. We want to do more. We want to have things be cheaper. And so we have all this demand and we're trying to keep up with our own society'…”
Gomez: Inner Monologues Are Crucial for Reliable AI Assistants
“I think the notion of using reasoning or letting the model have an inner monologue to work through problems, think through them make mistakes, but then realize that catch mistakes and correct them. I think that's a crucial piece in improving not only the accur…”
Gomez: AI Models Are Less Capable and More Controllable Than Feared
“As more and more evidence emerges that these models are much more controllable than we may have thought that they're a little bit less capable than we may have thought it's harder and harder to make that narrative.”
Gomez: AI Safety Debate Has Shifted From Doom to Practical Concerns
“I think you see the discourse shifting now. I think the discourse has begun to shift away from doom and existential risk. And now it's much more about practical concerns, which I'm really happy to see”
Gomez: LLMs Interpolate Known Skills But Do Not Generalize Beyond Them
“I think they can interpolate between skills, and so if they've seen how to do A, and they've seen how to do B, they can get kind of the average of A and B but they don't just go completely beyond anything that they've seen.”
Gomez: AI Self-Improvement Plateaus and Won't Cause Intelligence Explosion
“Not happening. It's not happening. It improves for, it can self-improve for a while, and then it tapers off. And so, yeah, you get some good improvement out of it, which is why we use it, but then it plateaus. It doesn't just keep going forever.”
Gomez: Cohere is hiring PhDs to train AI models
“We're kind of at that level where we're currently hiring PhDs to teach the model in their specific domain.”
Gomez: Synthetic data struggles outside verifiable domains like math
“Synthetic data Probably doesn't get us out of that, that issue. I actually, I don't know if synthetic data outside of easily verifiable domains like math, it's hard to use synthetic data to drive outcomes.”
Gomez: Synthetic Data Makes Up a Growing Portion of Cohere's Training
“More and more synthetic data is becoming a huge chunk of the data that we train on.”
Gomez: AI Can Reach AGI Through The Internet Without Physical Embodiment
“I actually take, I think the less popular view, which is The internet is enough, and by observation, you can actually learn enough to be extremely, extremely compelling. I think that's if we're talking about AGI and doing things as well as humans do, I think t…”
Gomez: Cohere Powers Over 50 Applications in Oracle Enterprise Suite
“So there's some good examples of that with our partner Oracle, which they have this suite of applications, which basically power enterprise, HR, supply chain all of these sorts of back office functions. And we're powering over 50 different applications within …”
Gomez: Real Generative AI Value Comes From Boring Enterprise Productivity
“Maybe this stuff feels banal. Maybe productivity feels boring compared to some of the hype of AI, but it is the value. This is what we're trying to build for.”
Gomez: Cohere Has Zero Examples of Enterprise Customers Replacing Staff
“I am not aware of it. I don't think I have any example of that happening. It's very assistive actually. So it's less about replacement. It's more about augmentation. Like at the moment, what everyone's building are tools to augment their workforce to make them…”
Gomez: Cohere Targets On-Premises Deployments for Regulated Industries
“So Cohere has had a long time focus on, on-prem as well, because for a lot of regulated industries, like finance and healthcare, a lot of that data doesn't actually go on the cloud.”
Gomez: RAG Creator Patrick Lewis Leads Cohere's RAG Efforts
“The guy who created RAG when he was at Meta is Patrick Lewis, and he leads our RAG efforts.”
Gomez: AI will become daily collaborative partners within two years
“In the next two years, I think we're going to start to see really compelling assistance. It won't just be little convenience functions or small features. It'll look a lot like a partner that you do work with. Someone that you interact with every single day and…”