Charles Lollette is a Rockstar Tech reporter who recently joined Business Insider. He recently knocked out several scoops on Scale AI. This is a startup that Meta is investing in. Founder Alexandr Wang is also part of Meta.
If you want to understand what Mark Zuckerberg is buying, Charles' coverage is a must-read. He broke the news about how Google used scale AI to chase ChatGpt and led an exclusive report on how large AI customers were robbing startups. There is also a surprising story about how to scale sensitive data about Google, Xai and Meta that anyone can see on the web.
I asked Charles what he thought of this unique deal.
Why is Zack paying so much?
If you want a state-of-the-art AI model, one way to do that is to rewrite millions of chatbot responses to humans and do many small tricks to make each AI answer painstaking. Our report shows that scale AI is specialized for this type of work. For example, what AI did to help Google catch up with CHATGPT. This kind of expertise is valuable to Zuck. Because he's about to release a new model that will blow everyone away. The lukewarm reception for Lama 4 clearly reaches him, and he is acting aggressively.
What is the future of scale AI at present, and how can it help Meta and Zach's AI ambitions? Or is this a massive acquisition of Alexandre Wang?
It was great to see the Google Pull Scale AI Project a few hours after Meta Investment was announced. Whether scale AI can recover such lost revenue is an open question. But I don't think Meta will just drive out I Peter and die. Meta currently has more direct access to scale AI to train modern models. Zuck needs a llama model that is either at the top of the AI benchmark or very close. Scale AI has a good idea of what works and what doesn't, the platform for making things happen. This covenant was not just about hiring a king. There was no attack on Alex, but there was probably a cheaper way to do it.
What did your scale AI report reveal about how AI models are being developed?
Training AI models requires a large amount of human labor. An important way to improve the model is through human feedback or reinforcement learning from RLHF. Remote, the largest platform for scale AI, has more than 240,000 human contributors doing this. Ironically, automation is a problem here. These humans often prefer to copy and paste answers from ChatGpt rather than give real human feedback. To do this, I noticed many examples of humans being banned from scale AI platforms.
What else should BI readers know about this situation?
Zack has placed a big bet on Scale AI. People are comparing this to metaverse, burning billions and not panning out yet. However, this could be considered a sensible move once Meta begins to rise in AI rankings and performs better on the llama model.
Sign up for BI's Tech Memo Newsletter here. Please contact me by email abarr@businessinsider.com.
