Hello world!
We have just launched https://t.co/bMefI5hEB5, a community for #MachineLearning#Makers and #Entrepreneurs.
Our aim is ti attract ML people with maker or entrepreneurial spirit to encourage the development of ML real world applications.
Sharing is appreciated 📢❤️
The World Cup is about to start and I want to share a great paper on football analytics
⚽️VAEP - Valuing actions by estimating probabilities
Problem: players are valued by simple statistics like goals + assists or xG and xA, which constitute less than 1% of their actions
Attention is all you need (but this time in terms of marketing, not model layers).
Stable Diffusion has become a perfect viral content generator on social media. This attention is starting to drive more money into the AI field, and so growth will be hugely accelerated.
STOP complaining about bad quality data if you're a data scientist.
Most likely you'll keep your job thanks to it.
With clean data:
- Develop a model
- Deploy it (+data pipeline)
- Add MLOps with auto retraining when needed
- Now what?
COVID taught data scientists a huge lesson: MLOps is key for reacting as quick as possible to unexpected future events.
Let's see now if we really learned from it.
Being an autonomous Machine Learning engineer is getting really hard over time.
Now it's not just about mastering ML and DL, or tabular + image + text data. There's a huge new skillset to learn.
Should we look for ML generalists or specialisation is the way to succeed?
With the rise of MLOps, software engineering skills are becoming even more important when working in data science. This will impact people switching careers from a less technical position, as well as bootcamps offering data science programs with low entry requirements.
One of the first things you learn when you start working as a data scientist in most companies.
You have small or middle sized tabular datasets, you try NNs because you just took some DL course and they look cooler, but then a "simple" model dot fit on Sklearn performs better.
Machine Learning in 2022:
- Everyone amazed by DALL-E 2 and thinking that AI will replace them on their jobs
VS
- 99% of data scientists still sweating every single day trying to create some value from the noisy tabular data they work with and bring it to production
I've tracked tweets talking about Clubhouse and their sentiment score from around 6200 online entrepreneurs.
- Hype is going down
- Haters percentage is almost constant
- Peaks are explained by big players entering the platform (@elonmusk , @naval ...)
I'm tracking the fastest growing Maker accounts in https://t.co/TXF4uwqNw9 and here are top ones for the last 30 days:
@rameshvel 172▶️878 (🔼410%)
@hrishiptweets 766▶️2381 (🔼210%)
@ryankramerllc 204▶️615 (🔼201%)
@basakbuilds 681▶️1773 (🔼160%)
@vponamariov 971▶️2509 (🔼158%)
Today I'm launching https://t.co/TXF4uwqNw9 🔥
🔍 Discover 5000+ online makers / indie hackers / entrepreneurs (+ updates)
⚡️Find their products
📑Filter them by Twitter bio, stats, topics they talk about
📈 Search fast growing accounts
... and many more to come 😀
Notebooks are a messy tool for #ML if you don't write reusable code. Invest time on creating your own AutoML framework and you'll work 10x faster.
Object oriented programming forces you being organised and thus more productive in the long term. Focus on SW engineering skills.
I loved @madewithml since it came out but I even like more how it has pivoted. @GokuMohandas perfectly expresses what I feel about the noisy abundance #MachineLearning content for Data Scientists / Sw devs. Filtering quality content is the right path.
https://t.co/z1jUAfjtZJ
Will we come back to the starting point? We didn't want to pay for individual songs or movies and then Spotify and Netflix changed our consumption behavior. Will paid content be aggregated under big platforms? If so, the abundance vs filtering content problem arises again.
5/5
That means that:
- Price competition will be huge. This value perception depending on price has a limit on how much money people are willing to spend in total on online content.
- Paid content will have less value perception as many people will try to make quick money.
4/5
The challenge now will be pricing content properly. Paying 10$ for reading a single newsletter is cheap, but of course people need more than one piece of valuable content per week. As this is becoming a trend, more paid content will be created.
3/5