- RuntheAI
- Posts
- LLMs Learn to Learn on Their Own 🔥
Good Morning AI Runners 🏃♂️
Here's what we've got for you today:
Google denies ChatGPT data used to train Bard
Microsoft's GPT-4 Blazes Trails as an Early AGI Prototype
LLMs Learn to Learn on Their Own 🔥
The latest AI drama comes from Google's Bard, which hasn't exactly had a roaring debut. So, to give it a boost, Google's DeepMind division is teaming up with Google Brain to launch Gemini. However, things get “spicy” as The Information alleges that Bard was trained using data from OpenAI's ChatGPT, which was scraped from ShareGPT. A former Google AI researcher, Jacob Devlin, supposedly left the company to join OpenAI after warning them that it would violate OpenAI's terms of service and result in similar-looking answers. However, Google is denying that they used this data, though they won't say if they used it in the past.
Microsoft Research just published a 154-page report titled "Sparks of Artificial Intelligence: Early Experiments With GPT-4," which suggests that GPT-4 could be viewed as a kind of early-stage proto-AGI. In the paper, the researchers report on their investigation of an early version of GPT-4 that can solve difficult tasks that span mathematics, coding, vision, medicine, law, psychology, and more, without needing any special prompting. They suggest that GPT-4's performance is strikingly close to human-level performance, and often surpasses prior models like ChatGPT.
LLMs (large language models) are already quite impressive. But what's even more impressive is that they can now improve themselves without needing any additional training data, reinforcement learning, or human intervention. This means that LLMs can generate an initial output. They can then analyze that output and provide feedback to themselves. Based on this feedback, they can improve their own output. And they can keep doing this without any external help, making them even more autonomous and powerful than before.
Pic of the day:
That's it from RunTheAI for today.
THANK YOU FOR READING AND SEE YOU TOMORROW, SUBSCRIBE TO STAY UPDATED!
P.S. if you made it this far, hit “reply” and tell me what you think of today's newsletter...what’d you love? What was boring?