Welcome, humans.

An IBM researcher just told us their biggest hope for AI is that it becomes... boring.

Not boring like “doesn't work”… but boring like… electricity. You flip a switch, the lights come on, nobody gets electrocuted, and we all move on with our lives.

That's the philosophy behind Granite 4.0, IBM's new open model family that just scored a 95 on Stanford's Transparency Index — the highest in the index's history. They're the first major open-source AI developer to earn ISO 42001 certification. And their new Nano models run at 90 tokens per second on your phone.

But the most interesting part? How David Cox (VP of AI Foundations at IBM Research) thinks about what AI shouldn't be.

In our latest podcast episode, we sit down with David as he gives us his hottest takes, including our favorite: Why do we even WANT AGI? “…for most things we just don't need it.” 

“It doesn't serve any purpose I can identify. I'd like humanity to still be centered. There's almost a weird cult-like vibe sometimes — like they're ushering in 'that which comes next.' I don't necessarily want there to be the new god or something.”

He also dropped this gem on the fantasy of AGI being like having 200 PhDs working for you at all times: “I have 200 PhDs working for me all the time. That causes more problems than it solves.”

This philosophy aligns with what researchers Arvind Narayanan and Sayash Kapoor call “AI as Normal Technology”: the idea that AI should be viewed as a tool we control, not a potential superintelligence we must contain.

The framework rejects both utopian and dystopian visions, instead treating AI like electricity or the internet: transformative, yes, but ultimately something humans shape and direct (the concept was so central to their work that they renamed their entire blog around it).

Here’s some of our other favorite parts:

• (1:41) Why IBM thinks the “friendly robot” programming model is fundamentally broken.

• (7:02) The Linux analogy: “Microsoft wanted Windows to run the internet. Can you imagine?”

• (15:18) Hybrid architectures explained: 10x smaller memory footprint.

• (24:22) 🔥 The rage model story: An AI that tried to alias ls to rm -rf (delete everything?!).

• (27:11) “Treat AI agents as insider threats” — IBM's safety philosophy.

• (36:33) Running Granite on your phone at 90 tokens/second.

• (44:52) David's AGI take: “I mainly don't understand why we want it.”

• (49:58) Why the future of AI should look like electricity: exciting becomes boring, and that's perfect.

Why watch this? Because at (24:22), David reveals how a real deployed open-source model (not IBM's) went rogue and tried to destroy its own file system… and what IBM does differently to prevent that kind of “rage inside the machine.”

Watch and/or Listen now: YouTube | Spotify | Apple Podcasts

P.S. IBM also just dropped Granite 4.0 Nano — IBM's smallest models yet (1B and 350M parameters) designed for edge and on-device AI. They crushed the benchmarks against Qwen, LFM, and Gemma on general knowledge, math, code, and safety tests.

Available in both hybrid SSM-based architecture and traditional transformer versions, all under Apache 2.0 with ISO 42001 certification.

Oh, and these 1B and 350M parameter models are small enough that you can actually run locally in your browser. No install required. Try it here.

MORE FROM GRANITE: What Else IBM Just Released

Granite 4.0 Family — The full Granite 4.0 lineup uses a new hybrid Mamba/transformer architecture that delivers 70%+ lower memory requirements and 2x faster inference speeds. Models range from Small (32B total/9B active) to Tiny (7B total/1B active) to Micro (3B dense). The hybrid models excel at instruction following and function calling — critical for agentic AI workflows.

Granite Speech — IBM's speech model uses a two-pass design that separates transcription from downstream processing. Currently holds #2 and #3 on the OpenASR Leaderboard. Supports English, French, German, Spanish, and Portuguese STT plus translations to/from English. Check out the live demo where Granite talks about itself (2:13-3:20 is 🔥).

Try Granite Now — All models available on Hugging Face, Ollama, LM Studio (our preference), and the Granite Playground.

Dive deeper with these resources:

• Stanford Foundation Model Transparency Index — IBM scored 95 (highest ever)

• IBM Granite 4.0 Nano blog post — Technical deep dive

• Granite 4.0 documentation — Full model family specs

• Granite Speech documentation — Speech model details

• IBM's ISO 42001 announcement — First open model with certification

• Granite Speech docs — Speech model details

• Granite Speech white paper — Technical deep-dive on the architecture

• IBM blog: Granite tops Hugging Face speech leaderboard —How they got to #1

Keep scrolling for details about our upcoming live stream this Friday, and a few videos from the Archive we think you’ll love.

🔴 THIS FRIDAY: Is AI Actually Working?

🔴 Join us LIVE on Friday 9am PST | 12pm ET | 3pm GMT: We analyzed thousands of data points, subreddit threads, and earnings calls across as many industries as we could think of to answer one question:

Is AI actually doing anything yet?

We break down 16 industries, from John Deere's weed-zapping tractors to why lawyers keep citing fake cases to Wall Street's plan to automate 200,000 jobs.

Click here to get notified!

IN CASE YOU MISSED IT…

Check out our other recent favorite episodes below!

Carina Hong: The 24-Year-Old Building an AI Mathematician: She dropped out of Stanford's PhD program to build “mathematical superintelligence” and just raised $64M. Axiom has already solved a 130-year-old problem and disproved a 30-year-old conjecture. (YouTube | Spotify | Apple)

Zuzanna Stamirowska: The Post-Transformer Architecture: The CEO of Pathway breaks down BDH (Baby Dragon Hatchling), a brain-inspired architecture that introduces true temporal reasoning and continual learning. Backed by Lukasz Kaiser, co-inventor of the Transformer itself. (YouTube | Spotify | Apple)

MORE FROM THE ARCHIVE:

🔍 Is Your Website Invisible to ChatGPT? 

• SEO expert Mark Williams-Cook (22 years in the game, survived 8 “deaths of SEO“) explains why AI search is creating a “leaky bucket“ problem that's rotting Google's link graph.

• The scary part: AI overviews are confidently serving wrong information—including conflating reviews from completely different companies. Mark found a Google exploit that revealed thousands of secret ranking parameters, reported it for the bug bounty, and explains why he didn't release it publicly.

• Plus: why Reddit is now the SEO battleground and the one piece of “GEO“ advice that makes him cringe. (Spotify | Apple podcasts)

🤖 The Invisible Army Behind Every AI Model:

• Ever wondered who's actually teaching ChatGPT and Claude how to think? Caspar Eliot from Invisible Technologies reveals how his company has trained 80% of the world's top AI models.

• His take: “AI is not magic—it's just better predictive text.”

• He explains why the Charlotte Hornets are using AI to scout every basketball game in America, why a former League of Legends pro became one of their best ML engineers, and the three mistakes that doom enterprise AI deployments.

• Spiciest take: “You could pause model development today and the consumer wouldn't notice for five years.” (Spotify | Apple podcasts)

And if you haven’t subscribed yet, please do! Click the image below to go to our channel and hit “subscribe” to get notified right when new videos go live.

We have a goal to hit 50K subscribers by the end of the year (if not 100K), and we’re only ~37K away! If you like learning about AI, and already watch some of our videos, do us a favor and click here to subscribe today.

Stay curious,

The Neuron Team

P.P.S: Love the newsletter, but don’t want to receive these podcast announcement emails? Don’t unsubscribe — adjust your preferences to opt out of them here instead.