OpenAI Shipmas Day 2: Huge Update that could lead to major breakthroughts - Model Customization with Reinforcement Fine Tuning. #openai #ainews #artificialintelligence #chatgpt #MerryShipmas #ai

OpenAI launched something that may seem minor at first glance, but it's actually huge. Especially if you're in research, healthcare, legal, or someone that just wants to customize your own large language model. So I'm gonna explain why today's update can completely change the future of a lot of really important fields and why it matters to you. I'm Celia, I make AI easy, so let's dive in. So today OpenAI announced a way for people to take their base models, fine tune them to their own specific needs. Especially those with really niche knowledge areas that requires a lot of highly specific knowledge that their base models may not be trained on. So things like law, genetics, research, that kind of stuff. And the best thing is you actually don't need a massive data set of your own in order to train your new model. Instead it could take a couple of dozen training examples. I'm not gonna get overly technical, but basically this will allow people to custom train their own models to reliably get more correct answers more regularly using something called reinforcement fine tuning. Basically by providing specific training examples and specific evaluation criteria, you can build a model that has much stronger reasoning for your specific niche use case. That was a lot, here's why it's a big deal. For example, in research for rare diseases or genetic diseases, which was the example on the live stream today, there can be a lot of limited data sets and a really strong need for deeper understanding. Being able to train a model on what we know so far and helping to intuit these different patterns across these different data sets and training data can be really impactful for people that are kind of overlooked in the research space because their disease is rare. So instead of relying on generic AI models, which may not be able to produce that kind of answer because it's not in their training set, you can customize something to your research and leverage your own expertise for that particular use case. And now, and it's not widely available yet, OpenAI did say that they plan to roll this out to be available for use sometime early next year. But in the meantime, they do have an alpha program that researchers and scientists can apply to. You can find out about it on their website. And they do have updates on their websites explaining this probably in a far better way than I just did. They do have 10 more days of shipments left. I am going to try and give an update with each day. So if you found this helpful, follow for more updates. Check out the one from yesterday, which I thought had more of an impact for the average OpenAI user. And check out my book, AI for Life, if you found this helpful.