Hey, . Welcome to the monthly version of the Artificialis newsletter!
A summary of what happened during the month, the best blogs of our Medium publication, latest news of our Discord server events, and recent news in the world of Artificial Intelligence and Machine Learning!
So sit back, take a cup of coffee, let's go over what happened this month.
A little bit from me and the server:
I started being a bit more serious on Twitter, trying to post daily. It'll be a mix of what you see on Discord, LinkedIn, and some short-form content. Make sure not to miss it!
Some new blogs:
- Building Music Recommendation Systems, by Sithu Khant
- Given the tip you'll see later in the newsletter, I'd like to re-purpose one of my old blogs - Maximizing Model Performance with Knowledge Distillation
AI in the world
*Small disclaimer: this issue will include both July and August hottest news in AI and the tech space
- Google started a competition to develop “Machine Un-Learning” technology, algorithms to make AI models “forget” rather than learn from specific datasets. Interesting approach to solve bias and, potentially, data privacy issues.
- Hong Kong-based biotech company InSilico Medicine has developed an AI-generated drug to treat lung disease. The drug is the first of its kind, with a novel target and design, created by artificial intelligence and is currently undergoing Phase II clinical trials.
- Anthropic has launched Claude 2, their improved text-generating AI model. Claude 2 surpasses its predecessor, offering enhanced performance in tasks such as coding, math problem-solving, and multiple-choice exams. It also comes with an expanded context window and updated training data!
- Llama-2, by Meta, is a game changer: it's better than any other open-source model, and its performance is at par with GPT-3.5 turbo. Although far from GPT-4, it's still impressive, and it's not over! The model is 100% free, open-source for research and commercial purposes. I always knew open-source was going to win.
- Elon Musk intends for Tesla and his new business, xAI, to work together on both the "silicon front" and the "AI software front." He also claimed that xAI will train its artificial intelligence systems and products using Twitter data.
- As user demand for these services increases, Apple plans to provide generative AI tools to keep up with rivals like Google and Microsoft. However, it appears that the corporation is giving privacy issues top priority. A "thoughtful basis" will be used to apply AI to more offerings, according to Apple CEO Tim Cook.
- AudioCraft, the generative AI framework created by Meta, has been made open-source. It consists of three models that take text inputs and produce audio, music, and sound effects. Researchers and professionals can now use the models to build their own models and train them.
Tip of the month
Best overfitting advice
This is the best advice I read on combatting overfitting:
“To achieve the perfect fit, you must first overfit”.
Here are the reasons why:
First of all, it makes sense; without an overfit model, you cannot combat overfitting.
Second, it is a sign of strength; if a model is overfitting or perfectly remembering the training data, it means the model has the power to optimize and can learn the patterns in the training data.
The tension between optimization (how well the model learns from training data) and generalization (how well the model performs on unseen data) is the key to solving ML problems.
When you have a model that can overfit, you should concentrate on generalization since too much optimization degrades the model. You ought to experiment with less intricate model architectures, use regularization, incorporate random dropout layers, etc.
'Till next month, you can find everyone here:
Have a fantastic month, !
If you'd like to support our community, and have access to millions of amazing articles and tutorials, consider subscribing to Medium's program via our link, we'll receive a small portion of your fee.
All of the money will be used to sponsor our Discord's events prizes!
0 comments