Jeremy Howard of fast.ai— The Simple but Profound Insight Behind Diffusion

Jeremy is also a co-founder of #Masks4All, a global volunteer organization founded in March 2020 that advocated for the public adoption of homemade face masks in order to help slow the spread of COVID-19. His Washington Post article "Simple DIY masks could help flatten the curve." went viral in late March/early April 2020, and is associated with the U.S CDC's change in guidance a few days later to recommend wearing masks in public.

In this episode, Jeremy explains how diffusion works and how individuals with limited compute budgets can engage meaningfully with large, state-of-the-art models. Then, as our first-ever repeat guest on Gradient Dissent, Jeremy revisits a previous conversation with Lukas on Python vs. Julia for machine learning.

Finally, Jeremy shares his perspective on the early days of COVID-19, and what his experience as one of the earliest and most high-profile advocates for widespread mask-wearing was like.

---

⏳ Timestamps:
0:00 Intro
1:06 Diffusion and generative models
14:40 Engaging with large models meaningfully
20:30 Jeremy's thoughts on Stable Diffusion and OpenAI
26:38 Prompt engineering and large language models
32:00 Revisiting Julia vs. Python
40:22 Jeremy's science advocacy during early COVID days
1:01:03 Researching how to improve children's education
1:07:43 The importance of executive buy-in
1:11:34 Outro
1:12:02 Bonus: Weights & Biases

---

📝 Links

---

---

💬 Host: Lukas Biewald
📹 Producers: Riley Fields, Angelica Pan

---

Subscribe and listen to Gradient Dissent today!

[dhruvpatel]

Jeremy is legend for us the small guys.

[AmputeerMeneer]

Jeremy Howard, the hero we need but don't deserve

[vaibhavnakrani]

Always love listening to Jeremy.
Feels like home!

[xaisthoj]

I'm learning to model ANN in the graph programming languages of Pure Data and Bespoke Synth. Way better to both see/understand and debug/experiment in than text programming languages.

[paultiplady]

Anyone know which new papers he was referring to in the "Diffusion and generative models" section?

[Silduril]

Huh, I actually enjoyed the second part of this interview (masks + education) more than the ML stuff... nice surprise 😄

[mikejones-vdfg]

Very cool, machine learning is sorely lacking in today tech which isnt surprising but leaving me a little impatient and upset when i see brand new tech that could really benefit from it but isnt like a 3d scanner review i was watching the other day (related to 3d printing which also is lacking any real AI to ensure accurate prints). In the review she took like a million images in 360 degrees and it couldnt really accuraly model the thing it was scanning... meanwhile i seen a nueral network that could render a 3d model out of just 1 2d image, seemingly with better results. The power is here but not being used yet which is kind of cool, its the wild west right now and I agree the spark of scientific evoltuion about to happen will be exciting and could have us exploring the universe sooner then we imagined.

[RR-etzp]

read the unplugged alpha by richard cooper

[marcoramponi]

25:49 "openai models that you can access are not the same as those described in their research papers". This is VERY important! Especially for the research community. Can someone provide some more context/evidence for this statement???

[ze-speeches]

I anticipated ChatGPT to be one of the discussion topics since the video was published in January 2023 😁It would be great to include the recording date in the title if possible. Otherwise, excellent conversation, thank you!

[billross]

What python also lacks: java-style multithreading.

I just realized I'm half of a diffuse psychology GAN! The other half asks exploratory, Rorschachian photo-pair 'questions', while I label and retrain. 224x224 pics boil down to two numbers per pic per side, accuracy 90%. New way of modeling personality.

I'm looking for a first open source (AGPL) victim to try it (add pics, label pairs, predict new pairs). Jump-start your inner iterative loop! Wide open ML field, mi muy eml is primitive.

[coloryvr]

Wow! Big Fanx for that Video, I am using SD to create textures and elements for my Virtual Reality Art Videos. ...and I am feeling like I woke up in a SF Movie . But I have one small concern: Will future generations unlearn how to think because the AI can do it better anyway...(?)
Finaly: I think the AI art will really unfold its true power in virtual reality...(I have heard SD is working on VR features(?)),