1: What is the best method for a beginner to get into ML, according to you? I know this question gets asked a lot.
A: I’m a bit biased, but I would say Introduction - Hugging Face NLP Course or https://www.youtube.com/watch?v=w18TCe8_XC0
2: Do you play Warframe, and are you named Clem after the iconic character?
A: I don’t play Warframe so I imagine the character is named after me
3: How is HF supported in langchain?
A: We’re very well integrated with them: Hugging Face | 🦜️🔗 Langchain
4: Favorite sci-fi book and/or movie?
A: I don’t read much sci-fi, especially on AI, because I feel like it’s very misleading when you’re actually building it. I feel like a lot of the “doomers” challenges come from too much exposure to sci-fi. I read a lot of philosophy, though, and my favorite book is The Myth of Sisyphus - Wikipedia
5: Do you have any plans to create tools similar to skops? ie tools that focus on classic data science, analysis and visualization?
A: Yes, we’d love that. Any specific existing repo we should look at?
6: Are there any plans to expand HF offices to other countries?
A: Yes, wherever there are Hugging Face team members (that we’re hiring all over the world), they can set up an office. Probably more than 15 by now!
7: What would you say is the best way to contribute to models adaptation/ development (for example, SDXL control-lora) and what channels could we use to coordinate?
A: Super excited about optimization & specialization for models! GitHub or the community tabs on HF are the best way to coordinate.
8: Clem, I don’t have a question, I just wanted to say, I think we met at the SD launch party in SF almost a year ago, and what an incredible year it’s been. HugggingFace is everywhere; the folks who are sharing on Twitter @Omar Sanseviero @VB @abubakar and @akhaliq are incredible for the community; the work you’re doing to help give the world open source, transformers (and transformers JS shoutout @Xenova ) are honestly for me lifechanging, I couldn’t have built Targum without them.
So, no question, keep hiring great folks, keep giving us all spaces to share open source models, compare them, and follow everything else that comes out Kudos on the raise, I’m sure it’ll go towards some A/H100s
9: What do you play?
A: Used to do a lot of Counter-Strike!
10: I imagine that both a larger vision you have and external (unexpected) forces impact how Hugging Face develops. Looking at the past as a baseline - how large has one been when compared to the other?
A: I’m a strong believer that in AI - given that it evolves faster than anything else - the number one skill for a company is to adapt and evolve rapidly, which is why we’re very much doers and builders versus theoretical strategy people.
11: Any plans for more in-depth audio examples/ tutorials? Also, any word on efforts to have more Tensorflow & ONNX based examples and code?
A: Yes! You should talk to Patrick or other members of this team!
12: I love Camus’s works. Do you follow an absurdist viewpoint generally?
A: Very much so!
13: What are the main goals of HF this year?
A: Train more ground-breaking models like Starcoder and Idefics, grow the open-source and the platform 10x, grow the revenue!
14: How does Hugging Face use machine learning systems inside the company? Do you have an impactful use case with LLMs already, for example?
A: An interesting one is an automatic translation of the community tabs of the hub!
15: What are some Open Source values that you think connect more with HF culture ?
A: Transparency, inclusiveness, asynchronous communication, fair reward for all are all very important for us and very much embodied by open-source, IMO
16: What regulations, if any, do you think should be placed on ML and AI, specifically in the realms of generative images, video, and voice for content creation?
A: Depends on the countries and specific topics, but super excited about watermarking and generated content detection. I testified at congress earlier this year, so maybe there’s something interesting there: https://twitter.com/ClementDelangue/status/1673348676478025730
17: I’m curious, what’s your favourite non-transformers HF library?
A: TRL these days!
18: What do you think about Africa’s role in the field of AI and training more models in French and African languages ?
A: We need more models, datasets, and credits for sure! Maybe this is a good starting point: Languages - Hugging Face
19: There was quite some buzz some time back online about regulating AI and how certain bills would be unfair to the open-source community, but I haven’t heard much since. What according to you, is the state of regulations on AI and where is it headed?
A: This was my testimony a few months ago: https://twitter.com/ClementDelangue/status/1673348676478025730. Hopefully, we’re going to end up with good regulation! Feels like policymakers are trying to learn at the moment.
20: What about the human workers in Africa labeling datasets for 2$/H? As an AI player, what do you think about it?
A: We’re trying to push for more fairness, for example, in the starcoder project where this was an important factor in our decision making on what platform to use for annotation/RLHF: https://twitter.com/TolokaAI/status/1684594696553725953
21: Are multi-domain experiencing (and so atypical) profiles sought in terms of recruitment?
22: Any thoughts on a potential migration of ML/AI Engineers from Python to Mojo? Do you think it will happen?
A: I haven’t looked at it too much, but it looks cool!
23: Is there interest in a “Profile README” and pinned models/datasets, like GitHub has, to improve user profiles on the Hub?
A: Yes, I feel like it could definitely be useful at some point!
24: Will there be more merch options releasing soon?
A: Just released a new summer hat! Resort Cap
25: Any plans for more HF coursework? Love the current units! Would love to see additional content on PEFT based techniques!
A: I don’t think there’s anything planned yet, but would be cool indeed. Have you checked the gradio course with Andrew Ng? It’s quite good!
26: Hi! You said you will mostly spend your time on recruitment next month on twitter, especially based on diversity. When can I expect a return from you? Thanks!
A: I got like a thousand messages + we had to rush the round announcement because it leaked so sorry for the delay. It’ll be my main focus next week, so probably then if my inbox doesn’t explode in the meantime!
27: You have arranged a couple of hackathons in the past… Speech to Text, Diffusers etc. They are a pretty good way of democratizing NLP to underrepresented domains/languages etc. Please continue to do that!
28: Is there anything we can do to encourage the stable growth of Hugging Face? Any tips for managing resources?
Where can we suggest ideas that might help improve the underlying efficiency of Huggingface Hub? Forking models requires doubling the storage consumed on the LFS servers, which feels impossible to sustain.
Is there a way to apply for inclusion in any research efforts? I’ve seen models trained by HF, and I would love to be a part of that effort.
A: We’re taking a very conservative approach to growing - just 170 team members when we could have 5x more. Keeping a 10-year runway to make sure we’re building for the long-term for the community!
29: Do you plan to come to France soon ?
A: I was there for two months and just got back to the US a week ago
30: So employees at HF get equity?
A: All of them!
31: Hi, @Clem !
Impressed by where you’ve led ! Where do you see LLM technology headed as hype fades and use cases emerge?
A: I’m excited to see more time series, recommender systems, biology, and chemistry. Lots of interesting use cases and positive impact to build there!
32: And do you invest in other open-source AI startups?
A: Not a lot because I don’t have a lot of free time, but sometimes, when I can help
33: Does Hugging Face invest in other AI startups ?
A: Not at the moment. We give credits from time to time to support though
34: Can you tell us more about how the company tends to let HF engineers communicate themselves about their work and team achievements on social media ?
A: We believe it’s part of everyone’s job at HF to interact with the community versus delegating to specific people. Everyone can tweet and post from HF’s social accounts for example!
35: Isn’t the point of an IPO the fact that it’s the first time the public is offered equity?
A: Public is the important word here. Before an IPO, you can exchange shares “privately”. After the IPO, you can exchange shares on the “public” markets.
36: Do you plan to build a dedicated Research Lab at HF? Or there is already one that we don’t know of?
A: We have one that has trained models like Bloom (with Big Science), Starcoder, Idefics and more!
37: Could you briefly talk about Hugging Face’s general approach to sustainability topics, like energy consumption and carbon emission?
A: Our main focus at the moment is to create awareness and transparency on the topic (because it feels like a lot of the builders don’t care). https://twitter.com/SashaMTL is leading our efforts on the topic with things like https://twitter.com/techreview/status/1592221789819613185 or https://twitter.com/SashaMTL/status/1640420970434314247?s=20
38: It’s great to get money. But if it comes from the AI oligarchy, it’s important the details of the arrangement. Will Huggingface work like Apache or more like GitHub? Are you now ultimately on the path to ownership and centralized control like so so many before you?
A: This is an important point!
There are no strings attached or commitments whatsoever with this investment, and we made sure to have so many of these big guys so that we don’t increase dependency on one of them. If anything, we are trying not to get influenced by big technology companies but rather to influence them to do more open-source and community AI. We’ve already managed to convince them to open over 1,000 models on the Hugging Face Hub, and we are going to try to keep convincing them to open-source more! It will take a very collaborative approach, in my opinion, to keep building and growing the AI field that we want (aka open, collaborative, inclusive and more).
39: Hi @Clem , are there any extension plans for Hugging Face in Eastern Asia? And will there be any potential cooperations with companies in Chinese mainland for example Tencent, AliBaba.
A: We already have team members and users all over Eastern Asia. Hopefully we can do more in the future!
40: Any chance of getting the Muse or Imagen weights released?
A: Would be awesome! You should ask the researchers and companies to do so!
41: Is there a plan for adding features for hosting more tabular models on the Hugging Face Hub?
A: Yes! What should we do?
42: Does this fundraising come with an expectation for future profits? If so, what’s the plan for that?
A: We’re a for-profit company, so we do have to generate revenue and profit. We actually already do thanks to Hugging Face – Pricing, and expect for this to grow in the coming years. The cool thing IMO is that the more revenue we generate, the more we can invest in open-source & free stuff for the community.
43: I have been holding out hope for you guys as a strong beacon for true decentration/democratization. The news was really getting me down lol
A: We’ll try our best not to disappoint, and please keep us accountable if/when we’re doing mistakes as we most certainly will.
44: Hi @Clem Congrats on your fundraising! I am curious how HF got its first investment years ago to begin this whole journey. Really appreciate the work HF is doing for open source!
A: We were building a tamagotchi AI at the time This is the announcement of our very first investment: Hugging Face wants to become your artificial BFF | TechCrunch
45: There should be a way to make a category list for different industry segments based on the relevance of the particular tool to that industry
A: We’re thinking about some sort of a “collection” feature on the hub to aggregate models/datadets/demos. More soon!
46: @Clem Is there a standard procedure for sending big-ish feature RFCs to HF transformers?
A: Opening an issue/pull request is the best first step!
47: How can we (the community) invest in Hugging Face?
A: Haha! The best way to get HF shares is to join the team so that you get stock options!
- Contributing to open-source
- Life at Hugging Face
- Growth statistics / usage metrics
- Other! (comment below)