Top Data Science and AI News: March 2023
March has been a hotbed of news when it comes to AI and data science. Everything from the introduction of GPT-4 to Google’s Bard large language model, to the FTC warning AI labs to not expatriate their claims, March has been one eventful month. So let’s take a look at the top news stories related to data science and AI!
Goldman Sachs Report
In a report by the investment giant, over 300 million jobs may be impacted by LLMs such as ChatGPT. Since the introduction of ChatGPT last November, there has been a great deal of discussion within the data science community, policy comminutes, and even local water coolers in offices across the globe on this topic. With that said, according to a report by Joseph Briggs and Devesh Kodnani, around two-thirds of jobs are expected to feel some sort of impact as artificial intelligence continues to enter our daily lives. But what that means exactly is still up in the air. One thing is sure, many within tech, such as Microsoft CEO Satya Nadella, are betting that AI will not be a replacement for humans, but an enhancer of existing abilities.
1,000 Technology Leaders and Researchers Call for Pause in AI Development
Though for many the rapid adoption of AI outside of data science and academia has been an interesting development, another worry has begun to creep into the minds of technology leaders and researchers. And that is, the impact of AI as LLMs continue to grow in power at a rapid pace. That’s why over 1,000 technology leaders and researchers have called for a pause on any LLM more powerful than GPT-4 for at least six months. If not done voluntarily, they’ve asked that the government force a pause on the entire industry. But why? Well according to the letter, they’re citing “profound risks to society and humanity.”
Will Review AI Competition
Taking the current administration’s call to review AI and how agencies can lend oversight, FTC Chair Lina Khan stated that the agency will be paying close attention to developments to ensure a fair marketplace for smaller firms. But it didn’t stop there. The Chair of the FTC also put AI labs on notice about the claims related to the abilities of their products. Khan stated, “Sometimes we see claims that are not fully vetted or not really reflecting how these technologies work…Developers of these tools can potentially be liable if technologies they are creating are effectively designed to deceive.”
Google Announces Open Access To Their Chatbot, Bard
The much-anticipated announcement of Google’s Bard becoming open to testing finally came to pass this month. Though not yet clear on what features or changes, if any, Bard has been compared to OpenAI’s ChatGPT, and the release of this LLM has been highly anticipated as OpenAI has had a lock on the market since introducing their model last November. With the introduction of Bard, Google is signaling that it has no intention of releasing market share to any competitors, especially Microsoft. One thing is certain, Google’s move is a major development in the LLM sphere as more tech giants, firms, and startups see AI-powered programs as the future of innovation.
PyTorch 2.0 Officially Released
In what will likely be a boon for the Python community that focuses on machine learning, PyTorch 2.0 was released a few weeks ago. The popular open-source ML framework saw improvements in torch.compile serving as the primary API for PyTorch 2.0, the inclusion of the scaled_dot_product_attention function in the torch.nn.functional module, the MPS backend, and functorch APIs in the torch.func module. This release was early anticipated by the data science community and from the looks of it, the new features and improvements did not disappoint.
FTC Attorney Warns Businesses of AI Claims
In a blog post back in late February, FTC Attorney Michael Atleson warned businesses that they should be cautious about making exaggerated or false claims about their AI products. Without mincing words, he made clear the agency’s concerns related to the overhype of capabilities and unsupported claims that could result in the deception of customers. He said in part, “Your performance claims would be deceptive if they lack scientific support or if they apply only to certain types of users or under certain conditions.”
OpenAI’s Summary of GPT-4’s Abilities
With the release of GPT-4, OpenAI also took the time to provide eager users with a summary of abilities as well as a few case studies from a few well-known organizations. But one of the striking improvements to GPT3.5 is the introduction of 4 being a multi-model. In short, the LLM will be able to work with text, images, and video — a pretty big leap from the text-only input users have been used to since November. Of the big names, Duolingo stands out due to how it will use GPT-4. That’s due to the goal of enchanting the language learning app’s offerings by adding an AI conversation partner. And anyone who has attempted to learn a language will tell you, having a conversational partner is one of the strongest tools a learner can use to master a new language.
US Chamber Calls for AI Regulation Framework
The largest trade association in the United States, The U.S. Chamber of Commerce, called on lawmakers to pay closer attention to artificial intelligence and begin working on regulatory frameworks. In a report by the group, their concerns are that AI shouldn’t be a gateway to harm economic growth or create a national security risk. To do this, they asked for the establishment of a “risk-based regulatory framework” to ensure the technology is deployed responsibly. This represents a significant shift in the group’s usual stance of being hesitant on calls for regulation.
OpenAI Launches Whisper ChatGPT APIs to Democratize Advanced NLP Technology
In a bid to help the democratization process of advanced NLP technology, OpenAI launched APIs for ChatGPT and Whisper. For companies and developers who lack the resources to create their own chat models, this is a significant step that allows for the reallocation of resources to their projects and away from costly infrastructure. In terms of the greater AI marketplace, this launch by OpenAI can help to accelerate the development of innovative new AI applications.
Introducing ChatLLaMA: An Open-Source ChatGPT-Like Training Process Using RLHF for More Efficient AI Assistant Development
At the beginning of March, ChatLLaMA was released. What makes this interesting is that it provides an open-source alternative to the traditional training process, and it’s clear that this is the goal for Nebuly. What ChatLLamA can also do is make it easier and more accessible for developers to create AI models that are tailored to their users’ specific needs related to their projects. Though still in the early stages of development, because of what it can do for the overall community there is a significant level of interest.
Meta AI Releases New Large Language Model LLaMA
Finally, the last major story that came out in March was Meta’s entry into the large language model market with the release of LLaMA. At 65 billion parameters, LLaMA is designed to assist AI researchers and practitioners in developing and deploying NLP models who may not have access to the infrastructure required to run these kinds of models in a bid to “further democratize” the field of AI. At last reporting, LLAMA is still being released under a non-commercial license with access being granted on a case-by-case basis to academic researchers, NGOs, laboratories, and governmental organizations globally.
Next steps
What a month for AI, especially for large language models with GPT-4, Google’s Bard, and Meta’s LLaMA all being released. What was the summer of AI in 2022, is looking to be the year of AI in 2023. And with everything that is going on, you should subscribe to our newsletter so you can keep up on the latest news from the AI world. And when you do, check out ODSC East, coming in May. There you can see for yourself the latest advancements in AI by those who are pushing the envelope. See how AI will shape our future in Boston.
Originally posted on OpenDataScience.com
Read more data science articles on OpenDataScience.com, including tutorials and guides from beginner to advanced levels! Subscribe to our weekly newsletter here and receive the latest news every Thursday. You can also get data science training on-demand wherever you are with our Ai+ Training platform. Subscribe to our fast-growing Medium Publication too, the ODSC Journal, and inquire about becoming a writer.
