Jan 22, 2020 · 22m · mad
NLP—The Most Important Field of ML // Clement Delangue, Hugging Face (FirstMark's Data Driven NYC)
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
In this Data Driven NYC presentation, Clément Delangue, Founder and CEO of Hugging Face, argues that Natural Language Processing (NLP) is the most critical domain in machine learning. He outlines recent technical breakthroughs in transfer learning, Hugging Face's open-source ecosystem, and the rapid deployment of state-of-the-art language models across major enterprises.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 2.4% of the talking time here. How this is scored →
Matt as informed peer 0.3 Guest teaching 1.0 Guest disagreement 0.3 Matt pushing back 0.2
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 21:04 Reframing competition with tech giants
Clement rejects the premise that startups must fight tech giants directly, asserting instead that community open source collaboration yields a superior product to closed corporate efforts.
Hardest push from Matt ▶ 14:17 Probing open source success factorsHost Matt Turck presses Clement to explain the underlying strategic reasons for Hugging Face's open-source traction beyond general market timing.
Biggest teaching moment ▶ 14:36 Bridging science and engineeringClement educates the room on the organizational divide between corporate research labs and engineering teams, explaining how Hugging Face serves as the critical bridge.
Matt holds his own ▶ 14:17 Framing open source growth leversMatt Turck demonstrates domain insight by contextualizing Hugging Face's rise within machine learning dynamics and asking for strategic lessons.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Clement Delangue's Background & Hugging Face Overview | 0 | 0 | 0 | 0 | Clement Delangue delivers an introductory monologue outlining his professional background and Hugging Face's funding and traction. As this is a solo presentation without host participation, host-side expertise and pushback are scored zero. | |
| Thesis: NLP as the Most Important Field of ML | 0 | 1 | 1 | 0 | Clement presents his core thesis that NLP is the most critical field in machine learning, engaging the audience with a show of hands. Because this segment is part of the guest's monologue, host scores remain zero. | |
| Natural Language in Enterprise & Business Operations | 0 | 1 | 0 | 0 | Clement explains how natural language underpins human activities and introduces transfer learning as the pivotal technical breakthrough. No host participation occurs during this presentation segment. | |
| Benchmark Progress: GLUE Benchmark Performance | 0 | 1 | 0 | 0 | Clement walks through benchmark results showing machine performance eclipsing human baselines on GLUE and SQUAD. The host is silent throughout this segment. | |
| Shift from Human Language Understanding to Algorithmic Execution | 0 | 1 | 0 | 0 | Clement outlines enterprise adoption and specific production use cases across classification, information extraction, and search. Host metrics are zero as the monologue continues uninterrupted. | |
| The Virtuous Flywheel of Usage and Science | 2 | 2 | 1 | 1 | Host Matt Turck opens the Q&A session by asking Clement to explain the drivers behind Hugging Face's open source adoption before passing the microphone to audience members. The interaction is friendly and collaborative, featuring informative responses on monetization and big tech competition. |
Statements from this episode (15)
Delangue: Hugging Face raised over $20M from Lux, Betaworks, and Kevin Durant
“We raised a bit more than twenty million dollars from Lux BetterWorks, so a lot of, like New York investors and some Silicon Valley investors like A Capital and some fun angels like Kevin Durant before he moved to New York and the hat.”
Delangue: Hugging Face built the most popular open-source NLP library
“We mostly known For having built the most popular open source NLP library, which is on GitHub, that is called Transformers.”
Delangue: Hugging Face cited in over 700 research papers over 15 months
“We've been lucky to have been mentioned in more than 700 research papers for the last 15 months”
Delangue: Over 1,000 companies including Apple and Bing use Hugging Face
“And we have more than a thousand companies using our library from medium-sized one, like Monzo, for example, the banking one, to really large ones like Bing or Apple.”
Delangue: NLP is the most important field of machine learning
“I believe that NLP today is the most important field of machine learning.”
Delangue: Transfer learning was the missing piece enabling deep learning in NLP
“The last kind of, like, missing piece to this puzzle for us is transfer learning with the reference paper on the subject that came out a bit more than two years ago called Attention is all you need. And as a kind of, like, high-level, simplistic way that's wha…”
Delangue: Transformer models destroyed previous scientific benchmarks in NLP
“And what these, kind of, like, new, kind of, models did is that they started to basically like completely destroy all the previous science benchmarks, and really, kind of, like generate this immense, kind of, like, progression on, on the science side.”
Delangue: NLP models surpassed human baseline on GLUE benchmark by June 2019
“The human baseline, which means having humans do exactly the same tasks is just below what we reached in, in June last, last year.”
Delangue: AI is ending human exclusivity over natural language processing
“We're really moving from a world where only humans could do natural language, which is why you have search large customer support, sales, community, communication departments in companies.”
Delangue: Pre-trained NLP models require only a thin software layer
“Most of the intelligence is in the models, and the software engineering layer on top of these models is really thin. Which basically led these models to go to production really, really fast, right?”
Delangue: Google called transformer adoption one of its biggest changes ever
“Google did a lot of PR a few months ago saying that moving to these new transformer models has been one of the most impactful changes that they'done over the last five years, if not since the beginning of Google.”
Delangue: Square powers its customer support chatbots using transformer models
“Square, for example, uses Transformers to power their customer support chatbots.”
Delangue: Every company will use NLP within three years or face disadvantage
“I believe that in in three years, every single company in the world will use NLP. And even more than that, I believe that in three years, a company that is not using NLP will be in a systematic disadvantage compared to a company that is using NLP, because it m…”
Delangue: Hugging Face generates zero revenue and has not focused on monetization
“We haven't had the constraints to work on monetization yet, so we're making zero revenue.”
Delangue: Collective open source can outperform tech giants like Google and Facebook
“We believe that actually if all the other companies than Facebook, Google, Apple, Amazon are actually all contributing to the same open source technology, they can actually end up with something way better than each one of these guys could build.”