A long article in WSJ on March 7 just described how Google missed the opportunity to release a chatbot similar to ChatGPT under excessive caution and self-restraint. The author believes that executives are hesitant to maintain technological development advantages and maintain ethical safety...
A day later, the Google Robotics team, Berlin Technical University, and Google Research proposed a new artificial intelligence model PaLM-E, which consists of a set of neural networks that can process different forms of data, such as images, sounds, text, and robot control Strategy, to realize the control and operation of the robot.
A joint paper by the three teams explains in detail how visual data can be used to enhance language processing. Researchers first use large-scale image and text datasets to train a deep learning model so that it can learn some general features and knowledge and then fine-tune it on specific tasks to improve its understanding of images and natural language. understanding of the relationships between them and playing a better role in tasks such as robot control. The "material" used to train the model includes image annotation, visual question answering, and image classification.
The author also cites some specific usage scenarios, including:
-In robot control tasks, PaLM-E can generate robot control policies based on natural language instructions and visual input.
-In the visual question answering task, PaLM-E can generate corresponding responses from images and natural language questions.
-In the robot arm movement operation task, PaLM-E can generate movement paths and operation sequences according to natural language instructions and the robot state.
-In a multimodal dialogue system including text, image, and sound, PaLM-E can understand the user's natural language input and generate corresponding responses.
According to the team, they once directed the robotic arm to precisely move the red block toward the target cup as instructed — even though there were only three coffee cups in the training data.
Google's scientific research capabilities and financial reserves are beyond doubt, but as they say, there is a huge gap between a research prototype and a reliable product that can be used safely every day. And Google has to be much more thoughtful about its AI launches than smaller startups.
As the WSJ reports, two Google researchers began pushing the company more than two years ago to release a chatbot that could confidently discuss philosophical questions, talk about their favorite TV shows, and improvise about cows and horses. Puns aside, its technology was more advanced than any other chatbot at the time.
This was repeatedly rejected by Google executives because it did not meet the company's ethical standards for AI safety and fairness. Two researchers, Daniel and Norm, leave to start their own AI company. Google's recent emergency release of Bard, a chatbot, is based in part on the duo's technology.
Google's concerns about promoting the application of artificial intelligence technology mainly come from two aspects. One is that the long-standing controversy about artificial intelligence may affect the company's reputation, and the other is that it is worried that it will affect its advertising business.
The first worry has long existed in academia and technology applications. Even OpenAI was founded in the hope that it would one day be able to counter evil AIs that could be misused by large corporations such as Google.
In 2016, Microsoft released a chatbot named Tay, but only a few hours after it went live, some users "tuned" and published such as "Bush is behind the '9•11' incident" and "Hitler a bit That's right" and other sensational remarks.
The second is the unavoidable disease of large companies, that is, swinging between maintaining the existing profit model and customer base, maintaining stable cash flow and market position, and continuing to innovate in response to market changes and threats from emerging competitors. Usually in the end, the former defeated the latter, thus giving the latecomer a chance. It is one of the laws of business, and it is also the charm of the business world.