After more than half a decade Microsoft really huge disasterThe incident remains a stark reminder of how quickly AI can be destroyed after exposure to the powerful toxicity of the Internet and a warning against building bots without strong enough behavioral constraints. On Friday, Meta’s AI research department will see if its latest iteration of Blenderbot AI can stand up to the horrors of internet networks with the public beta of Blenderbot 3 of 175 billion variants.
One of the main obstacles currently facing chatbot technology (as well as the natural language processing algorithms that drive it) is sourcing. Traditionally, chatbots are trained in highly structured environments – because otherwise you always get TY – but this ends up limiting the topics they can discuss to the specific ones available in the lab. Conversely, you can have a chatbot pull information from the internet to access a wide variety of topics, but it will likely become completely Nazi at some point.
“Researchers cannot anticipate or simulate every conversation scenario in search settings alone,” Meta AI researchers wrote in a blog post on Friday. “The field of artificial intelligence is still far from smart AI systems that can understand us, interact and chat with us as other humans do. In order to build models that are more adaptable to real-world environments, chatbots need to learn from a diverse, broad perspective with “people” in the wild.”
Meta has been working on the issue since then First introduced BlenderBot 1 chat app in 2020. Initially it was little more than an open source NLP experience, by the following year BlenderBot 2 had learned to remember information it had discussed in previous conversations and how to search the internet for additional details on a particular topic. BlenderBot 3 takes these capabilities a step further by not only evaluating the data it pulls from the web but also the people it talks to.
When a user scores an unsatisfactory response from the system—currently hovering around 0.16 percent of all training responses—Meta returns feedback from the user to the form to avoid repeating the error. The system also uses a manager’s algorithm which first generates a response using training data, and then runs the response through a classifier to check if it fits a user-defined scale of true-false knowledge.
“To generate a sentence, the language modeling mechanisms and the classifier must agree,” the team wrote. “By using data indicating good and bad responses, we can train the classifier to punish low-quality, toxic, contradictory, or repetitive, and generally unhelpful, statements.” The system also uses a separate user weighting algorithm to detect unreliable or ill-intented responses from a human speaker – essentially teaching the system not to trust what that person has to say.
Our live and interactive Overview BlenderBot 3 enables it to learn from organic interactions with all kinds of people,” the team wrote. We encourage adults in the United States to try the demo, have natural conversations about topics of interest, and share their responses to help advance the research. “
The BB3 is expected to speak more naturally and talkatively than its predecessor, thanks in part to its widespread promotion. Language model OPT-175B, which stands approximately 60 times larger than the BB2 model. “We found that, compared to BlenderBot 2, BlenderBot 3 provides a 31 percent improvement in the overall rating of conversational tasks, as assessed by human judgments,” the team said. “It is also judged as poor knowledge, while factually incorrect 47 percent of the time. Compared with GPT3, in objective questions, it was found to be more current 82 percent of the time and more specific 76 percent of the time.”
All products recommended by Engadget are handpicked by our editorial team, independently of the parent company. Some of our stories include affiliate links. If you buy something through one of these links, we may earn an affiliate commission.