Click image above to go straight to YouTube and watch!
Welcome, humans.
What if AI could actually PREDICT what happens next?
Here’s what it can do today: ChatGPT can summarize a spreadsheet. Claude Cowork can run its own code to analyze your financial docs. Both Codex and Cowork can edit spreadsheet files.
But can these “language” based models reliably tell you what happens next?
That gap between explaining data and predicting from it is the whole argument behind our newest podcast. Neuralk CEO Alexandre Pasquiou says language models flatten away critical structure in business data, while tabular foundation models are built to learn from rows, columns, distributions, and numbers directly.
In our latest podcast episode, Grant asks whether we are wasting billions trying to make LLMs do a job they structurally struggle with, how Neuralk's Seldon model plugs into Claude, ChatGPT, and Excel, and why Alexandre thinks tabular models could become the predictive brain behind every company in 2-3 years.
Click the image to watch on YouTube.
Watch and/or listen now: YouTube | Spotify | Apple Podcasts
The key distinction between a large language model (or LLM) and a large tabular model is simple: LLMs are excellent interfaces and orchestrators, but Alexandre argues they are poorly matched to prediction on structured business data, where the relationships between rows, columns, and numeric distributions matter.
Seldon is Neuralk's bet on a different architecture: one general predictive model that can adapt to many datasets instead of forcing companies to build a separate custom analytical models for churn, fraud, demand, pricing, and every other forecast one needs to do in business.
Here’s our favorite parts:
(10:26) Summarizing data is not predicting from it: Alexandre explains why ChatGPT can query or describe a table but still struggle to learn a new dataset's distribution and make predictions at scale.
(22:25) Are we wasting billions scaling the wrong AI? Grant puts Alexandre on the spot, and he argues that bigger language models still do not fix the underlying mismatch with tabular prediction.
(28:30) Use predictive AI inside Claude or Excel: Alexandre walks through how Seldon's MCP, skills, and Excel integration let people send business data to a predictive model without building a custom ML pipeline.
(30:30) 20 million rows, 600+ columns: Seldon was designed to process tabular datasets far beyond a normal LLM context window.
(41:43) The 2030 prediction: Alexandre predicts tabular foundation models will power essentially every predictive workload, while agents automate most enterprise workflows.
The bigger idea is a multimodel future: you may keep talking to Claude or ChatGPT as your work partner, but those systems call specialized models behind the scenes when the job requires vision, prediction, forecasting, or another modality.
Why watch this? Because this episode explains a blind spot hiding in plain sight. If you use AI on spreadsheets, forecasts, customer data, finance, or operations, it shows why the best model for talking about your data may not be the best model for predicting from it.
Watch / listen now: YouTube | Spotify | Apple Podcasts
P.S. Jump to 33:09 for the wonderfully weird question: what happens if every company eventually uses the same predictive model?
Keep scrolling for tomorrow's live tool roundup, a practical Seldon explainer, and four recent Neuron conversations worth watching next.
🔴 LIVE TOMORROW: The Week’s AI Tool Roundup, But for Normal People
Thursday, August 20 at 10 AM PT / 1 PM ET, we're going LIVE to translate the latest launches into plain English.
The theme: what these new AI tools actually are, who they're for, and when you might realistically use them.
This is not going to be three developers yelling model benchmarks at each other for an hour. Instead, we’ll share whats new and why you, fellow normie, should care.
Qwen 3.8: what an open model is, why you might run one instead of ChatGPT or Claude, and when that makes sense.
Unsloth Studio: how to run and experiment with AI models on your own computer, even if you've never touched a terminal.
Cursor Origin: why Cursor suddenly wants to host your code too, and what that could mean if you build websites, apps, or internal tools with AI.
DeepSeek Harness: what an “agent harness” is, why people keep talking about them, and whether it matters outside hardcore coding circles.
Plus the other notable models and tools that dropped this week, and which ones are actually worth remembering.
And yes, if OpenAI drops Astra before we go live, we'll cover that too. The vagueposters have certainly been vagueposting, while OpenAI has publicly discussed slowing parts of frontier-model training to tighten safeguards.
If Astra arrives and turns out to be incredible, great. If it belongs in the “cool, another model” bucket... well, that is technically still part of the roundup.
The goal is simple: by the end, you should know what changed this week, what's useful, what's hype, and which tools are actually worth trying for your own work.
Bring your questions. No question is too stupid, and if the question is too smart, we'll ask the smartest AI we have access to for help. That's right, I'll waste a Fable prompt for y'all. You're welcome!
Additional Resources: What Seldon actually changes
Today's episode makes a useful distinction: LLMs can be the interface, while a specialized model does real prediction work. Neuralk built Seldon around that idea.
Oh, and anyone can use this right now… for free (to start at least! Hosted on-prem versions are available, too).
Use it where you already work: Alexandre says Seldon connects through Python, Excel, MCP servers, and skills, so Claude, ChatGPT, or another agent can hand off predictive jobs.
Skip the one-model-per-question treadmill: the goal is one foundation model that adapts across churn, fraud, demand, pricing, classification, regression, and forecasting problems.
Handle genuinely huge tables: Alexandre says Seldon scales to roughly 20M rows and more than 600 columns… that’s a lotta context y’all.
Keep sensitive deployments private: mission-critical companies can deploy Seldon in their own cloud or on-premises.
Explore it yourself: Neuralk / Seldon
🎙️ In Case You Missed It…
1. AI can write DNA now. Here’s what that means for the future of AI being used to “cure all diseases”
Click to watch on YouTube.
TL;DW: Radical Numerics CEO Eric Nguyen explains how genomic AI can read and write DNA, including complete viral genomes, while pushing toward multimodal models that combine DNA, RNA, proteins, and other biological signals.
Why you should watch: It makes the leap from “AI analyzes biology” to “AI designs biology” concrete, including the medical upside and the security problems that come with it.
Watch / Listen: YouTube | Spotify | Apple Podcasts
2. Want to run powerful AI without sending everything to the cloud?
Click to watch on YouTube.
TL;DW: Intel’s Dr. Olena Zhu explains why the future of AI may be hybrid: private and repetitive work stays local, while harder reasoning gets routed to bigger cloud models. Plus, she shares a staggering fact: if current trends hold, we might have Fable-class AI on our powerful laptops “within two years.”
Why you should watch: It turns “local AI” from a privacy slogan into a practical architecture for agents, cost, and reliability, along with the tools you can use to do it.
Watch / Listen: YouTube | Spotify | Apple Podcasts
3. Building something with AI? Watch: AWS Put a CTO Inside Claude Code
Click to watch on YouTube.
TL;DW: AWS startup leader Deap Ubhi explains how AI compressed startup iteration from months into days, while security, infrastructure, and reliability still separate a prototype from a business.
Why you should watch: It shows when builders should move fast and when technical shortcuts become expensive traps.
Watch / Listen: YouTube | Spotify | Apple Podcasts
4. How do you make truly autonomous surgery trustworthy?
Click to watch on YouTube.
TL;DW: Mathias Unberath explains why autonomous surgery is difficult, how developers test rare failures, and what reliability means when mistakes have physical consequences.
Why you should watch: It is a sharp guide to the gap between a technical demo and a dependable real-world system.
Watch / Listen: YouTube | Spotify | Apple Podcasts
New episodes of The Neuron: AI Explained explore the breakthroughs, businesses, and people shaping artificial intelligence.
Subscribe on YouTube by clicking below to help us get even more amazing guests like this one to teach you new things about AI every week!
Click this image to subscribe to our channel!
Stay curious,
The Neuron Team
What do you want to learn about AI?Pick your favorite, then share any others in the "additional feedback" |
|
BTW, we do read these answers! If you requested something from this list, we’re working on it! Write in with additional ideas in additional feedback after you vote.
P.P.S: Love the newsletter, but don’t want these podcast announcement emails? Don’t unsubscribe.
