So Nerdy Planet #3

Hello, Amazing Nerds! 

What a week, right? Sounds like a perfect moment to discuss news happening in the industry. 

So, today we will speak about:

1. Open-weight models, and why the world competes on those
2. GPT-5 Released. Is it good?
3. Unlucky Journey of Windsurf employees
4. Our Weekly Book Recommendations
5. So Nerdy Nation news!

1. The world of open-weight models

Before proceeding, we need to understand the first difference between open-source and open-weighted models since these terms are often used interchangeably. 

Open-source models are provided with everything: weights, code, architecture, and training process—sometimes even training data. Users can retrain models, audit them for bias or security, or even repurpose them.

Open-weight models provide only the final result of model training. Training code, dataset, or even architecture sometimes remain proprietary. Users can download weights and play with models. Some models provide capabilities for fine-tuning, but obviously, users can't retrain models. 

Nowadays in the market, there are almost NO truly open-source models, and there are a couple of reasons behind this ( good, bad, and ugly ):

1. (Good) Regulatory and ethical concerns—releasing a very powerful model as open-source will almost 100% guarantee misuse because that is what humanity does best.
2. (Good) Open-weight models satisfy most experimentation/research use cases while minimizing risks
3. (Good) Enormous cost and resources. Running models is extremely costly and, for most advanced models, requires extremely powerful hardware. Training state-of-the-art LLMs requires millions of dollars only in computer infrastructure, but also needs highly specialized talents ( and almost all of them under Zuck's aim ).
4. (Bad) Competitive advantage. Leading AI companies view the best LLMs as core intellectual property and business differentiators. Open-sourcing them would give away their lead to competitors.
5. (Ugly) Data licensing and copyright risks. We all hear about all these lawsuits filed by authors, artists, and even adult entertainment companies against leading AI Companies because they stole/scraped/pirated data for training models. Open-sourcing datasets will make them lose all these cases.