John Schulman has been tweeting about a variety of topics lately. In one tweet, he shares a short story by Isaac Asimov called "Someday" which features a language model called Bard. He also mentions fine-tuning the language model on recent data. Another tweet discusses the useful software skills for machine learning, highlighting the importance of front-end development in the era of pretrained models.
John Schulman also shares his excitement about his work on overoptimization of reward models and provides a link to a paper on this topic. He mentions getting access to @Cruise driverless ride service and being impressed with its performance. Additionally, he retweets an episode featuring OpenAI co-founder and inventor of PPO/TRPO, where they discuss RL from human feedback and AI alignment.
In terms of new trends in AI, John Schulman's tweets mention pretrained models, fine-tuning language models, RL from human feedback, and AGI timelines. These topics suggest that he is interested in advancements related to language models, reinforcement learning, and artificial general intelligence.
Overall, based on sentiment analysis of John Schulman's recent tweets, it is difficult to determine whether he is positive or negative about the way AI is going. His tweets cover various topics without expressing explicit positive or negative sentiments towards AI as a whole.
SOME AI BOOK RECOMMENDATIONS