#55. 🤖 AI-ONLY SPECIAL! | 💬 ChatGPT iOS app with Whisper voice-to-text | 🧐 Microsoft claims GPT4 has common sense like humans | 🛢️ Shell using AI to boost deep sea output
Plus: OpenAI channel for GPT prompt sharing | Are LLMs faking emergent properties? | GPT4 no big deal
⚠️ Yes, today's email is different... please read this:
So you'll be getting a simple yes/no survey about this on Saturday, but I think I might be changing Future Perfect to an entirely AI-only newsletter.
While topics like climate change and clean energy and autonomous/electric vehicles are dear to my heart, the data I'm seeing is pretty clear: AI is pretty much the only news that matters to you since I rebooted this newsletter to incorporate AI just two weeks ago!
So today's newsletter will attempt an AI-only format -- I'm thinking 6 articles + 3 cool links? — but please be sure to let me know at the bottom of this email what you think.
🤖 ARTIFICIAL INTELLIGENCE TOP NEWS
OpenAI has launched a ChatGPT app for iOS in the US, with an Android version coming soon. The app features numerous improvements over the web-based ChatGPT portal including the ability to search conversation threads and to speak via OpenAI's own Whisper voice-to-text technology that offers fluent and organic human-like voice input.
This OpenAI channel is full of people sharing GPT prompts for everything from applying to jobs to learning a language (read more; quick read)
Users of OpenAI's ChatGPT are sharing tips on a channel for creating refined prompts to perform a wide range of tasks. These include preparing for job interviews, learning new skills, and producing content that could outsmart AI detectors. Additionally, users share prompts to assist with creative work, such as writing fiction or short films, and for developers to enhance their coding and experiment with different programming languages.
A study by Stanford University researchers has suggested that the perceived "emergent abilities" of large language models (LLMs) like GPT-4 may be overhyped or misunderstood. While many believe that these models develop new abilities as they grow larger, the Stanford researchers argue that these "abilities" may not be inherent features of model scaling but instead a result of the metrics used to assess the models. Specifically, metrics that non-linearly scale a model's per-token error rate can give the illusion of emergent abilities. They also found that having a larger test dataset can make performance improvements appear more gradual rather than sudden. These findings emphasize the need for careful and critical analysis of LLM performance, suggesting that a better understanding of model scaling could lead to more responsible deployment and exploration of alternatives to creating larger models.
Rodney Brooks, a renowned roboticist, doesn't believe that models like GPT-4 would lead to artificial general intelligence (AGI) in the foreseeable future, as they lack underlying models of the world. He acknowledges the improvements they bring to tasks such as language processing but warns about their limitations and their tendency to give confident but potentially incorrect answers. He is skeptical of the massive valuations of AI companies and dismisses the hype around AGI, stating it leads to premature predictions about job losses or the arrival of self-driving cars. Brooks is optimistic about the use of AI in labeling images for robotics, asserting that these applications are already here but take time to deploy. He identifies opportunities in warehouse automation and advises young researchers to work on foundational issues in robotics, such as manipulation and touch, rather than focusing solely on machine learning.
Contrary to Rodney Brooks' viewpoint, a Microsoft research team has suggested that GPT-4 shows indications of artificial general intelligence (AGI). After having early access to ChatGPT-4, the team conducted studies and published a lengthy report asserting that the language model can be trained to reason and use common sense akin to human capabilities. This claim represents a significant development in the artificial intelligence field.
Shell is collaborating with analytics firm SparkCognition to leverage their generative AI technology to enhance deep sea oil exploration and production. This initiative aims to speed up the imaging and exploration of subsurface structures, reduce costs, and streamline the oil discovery process. Traditional methods require extensive data and complex physics-based algorithms, but with AI, seismic data will be used to increase efficiency and productivity. Both companies believe these innovative strategies will contribute significantly to meeting growing energy needs while minimizing the carbon footprint.
🤖🔗🤯 COOL AI TOOLS, APPS, VIDEOS, PODCASTS, LINKS, AND MORE!
🔥 Free OpenAI prompt engineering course: Grab it while you still can. In partnership with DeepLearning.ai.
💰 Alpha: AI-powered investing by Public.com.
💻 Tome: Another AI-powered slide generator.
🎉 THAT’S ALL FOR TODAY! PLEASE DON'T FORGET TO CAST YOUR VOTE BELOW!!
What'd you think of today's email?
This helps me improve it for you!
✅ Share this post if you liked it
✅ Share this link (https://www.futureperfect.news) if you're a fan of Future Perfect
✅ Vote below before leaving!
Thanks for reading!
Looking for past newsletters? You can find them all here.