1/ @huggingface: The #ArtificialIntelligence community building the future
Behind the hugging face emoji 🤗 is the fastest growing open source community in history
If you have not heard of @huggingface, it may be one of most important platforms of the next decade
Behind the hugging face emoji 🤗 is the fastest growing open source community in history
If you have not heard of @huggingface, it may be one of most important platforms of the next decade
2/ Founders @ClementDelangue , @Thom_Wolf, and @julien_c originally set out to build open domain AI.
Quickly, they realized the technology and platform they were building was bringing value to companies.
They decided to open source it
Quickly, they realized the technology and platform they were building was bringing value to companies.
They decided to open source it
3/ Two and a half years later, the platform has exploded to over 10,000 models, 1,000 datasets, and 45,000 GitHub stars
https://twitter.com/clementdelangue/status/1385349868428869636?s=21
4/ Why did that happen?
Artificial Intelligence and Machine Learning felt distant to most, but @OpenAI's release of #GPT3 shocked many
AI and ML is growing rapidly, and Natural Language Processing (NLP) like #GPT3 is a good example of innovations that can change industries
Artificial Intelligence and Machine Learning felt distant to most, but @OpenAI's release of #GPT3 shocked many
AI and ML is growing rapidly, and Natural Language Processing (NLP) like #GPT3 is a good example of innovations that can change industries
5/ What is NLP?
@IBM defines NLP as "the branch of computer science - and more specifically... AI - concerned with giving computers the ability to understand text and spoken words in much the same way human beings can"
@IBM defines NLP as "the branch of computer science - and more specifically... AI - concerned with giving computers the ability to understand text and spoken words in much the same way human beings can"
6/ Basically, it is a computer model that can understand us.
This includes sentiment (e.g., angry or sad), idioms (e.g., break a leg), confusing spelling (e.g., your vs. you're), etc.
Examples include Google Search auto-complete, iMessage suggested words, Siri voice recognition
This includes sentiment (e.g., angry or sad), idioms (e.g., break a leg), confusing spelling (e.g., your vs. you're), etc.
Examples include Google Search auto-complete, iMessage suggested words, Siri voice recognition
7/ How?
In a crude simplification, scientists / engineers build a predictive model on massive datasets
The model:
1. Tokenizes - Segments the data into digestable parts
2. Trains - Identifies patterns
3. Applies - predicts future patterns
In a crude simplification, scientists / engineers build a predictive model on massive datasets
The model:
1. Tokenizes - Segments the data into digestable parts
2. Trains - Identifies patterns
3. Applies - predicts future patterns
8/ Consider this example below with "Dogs are better than cats"
While this may seem simple, consider this: GPT-3 was trained on 175 BILLION parameters
While this may seem simple, consider this: GPT-3 was trained on 175 BILLION parameters
9/ So why does this matter?
First, these models are powerful and have many applications. Companies like @copy_ai are building on these.
Even so, NLP models are just one version. There are many more.
Second, these models are difficult to develop, understand, and operationalize
First, these models are powerful and have many applications. Companies like @copy_ai are building on these.
Even so, NLP models are just one version. There are many more.
Second, these models are difficult to develop, understand, and operationalize
11/ I further simplified into three categories:
12/ The applications are already wide, but they are expanding rapidly.
Recently, @huggingface was leveraged for protein language modeling:
Recently, @huggingface was leveraged for protein language modeling:
https://twitter.com/clementdelangue/status/1390003976045842436?s=21
13/ Challenges
There are, however, challenges. Scientists develop the models. Engineers often operationalize and commercialize them.
Currently, the industry is siloed, creating many challenges for both parties
There are, however, challenges. Scientists develop the models. Engineers often operationalize and commercialize them.
Currently, the industry is siloed, creating many challenges for both parties
14/ This is why @huggingface is critical to the industry.
Their open source platforms has connected science and engineers, centralizing a fragmented industry into a collaborative, innovative community
Their open source platforms has connected science and engineers, centralizing a fragmented industry into a collaborative, innovative community
15/ Scientists can now quickly publish models to thousands of engineers, receiving feedback and usage critical to further development
Engineers can quickly find, understand, deploy, and train models now. They are no longer wasting hours searching for fragmented resources
Engineers can quickly find, understand, deploy, and train models now. They are no longer wasting hours searching for fragmented resources
16/ @huggingface makes it as easy as one click...
https://twitter.com/julien_c/status/1390967581318819840?s=21
17/ The result has been a flywheel, accelerating both the development and adoption (or embrace… get it? Hugging?) of AI and ML models
18/ Conclusion
@huggingface is strategically positioned at the center of one of the most important fields of development of the next ten years.
It will truly become the GitHub of Machine Learning
@huggingface is strategically positioned at the center of one of the most important fields of development of the next ten years.
It will truly become the GitHub of Machine Learning
19/ The full writeup can be found on my Substack. I send ~1 a month
jeffburke.substack.com/p/hugging-face…
jeffburke.substack.com/p/hugging-face…
20/ Business Model
@huggingface is built on a tiered subscription (SaaS) model.
Subscribers get access to increasingly helpful features for model operationalization
huggingface.co/pricing
@huggingface is built on a tiered subscription (SaaS) model.
Subscribers get access to increasingly helpful features for model operationalization
huggingface.co/pricing
21/ Market sizing
Based on this, I have done a bottom-up analysis of market sizing based on regressive adoption of the tiered SaaS offering
Based on this, I have done a bottom-up analysis of market sizing based on regressive adoption of the tiered SaaS offering
22/ Investors
@Lux_Capital - @BrandonReeves08 & @wolfejosh
@svangel - @RonConway
@betaworksVC
@KDTrey5
@RichardSocher
@richkleiman
@fdouetteau
@Lux_Capital - @BrandonReeves08 & @wolfejosh
@svangel - @RonConway
@betaworksVC
@KDTrey5
@RichardSocher
@richkleiman
@fdouetteau
I forgot to add Lee Fixel to the list! I will tag the fundraise announcement here to avoid anymore!
https://twitter.com/ClementDelangue/status/1370044241603399682?s=20
• • •
Missing some Tweet in this thread? You can try to
force a refresh
