At the beginning of this year, Google introduced AI Test Kitchen, a app of Android that allows users to talk to one of their chat bots artificial intelligence more advanced, LaMDA 2.
Today, the company is opening registrations for early access. People who wish to participate can enter this link to sign up, and Google says it will soon allow anyone to download the app and start chatting. (Though it’s limited to US users at the moment.)
“We have been testing LaMDA internally for the last year, which has produced significant improvements in quality. More recently, we’ve done dedicated rounds of contradictory testing to find additional flaws in the model. We recruited savvy members of the Red Team (product experts who intentionally stress test a system with a contradictory mindset) who have discovered additional harmful, yet subtle, exits.” mention google on his official blog.
AI Test Kitchen will be home to a “rotating set of experimental demos” and three are currently available. They are based on the “latest version of LaMDA, which has undergone key security improvements”:
– imagine it: Name a place and LaMDA will offer paths to explore the imagination.
– List it: Name a goal or topic and you’ll see that LaMDA can break it down into multiple sub-task lists.
-Talk About It (Dogs Edition): continue the conversation and enjoy where it goes. It’s just a fun chat, a bit weird and open.
“Multiple layers of protection” are touted, including “techniques to keep conversations on topic, acting as guardrails for a technology that can generate endless and fluid dialogue.”
The artificial intelligences of Google and Meta, user-proof
Interestingly, when Goal made an almost identical move earlier this month, opening up its much more fully-fledged AI chatbot called Blender Bot 3to customers.
Sure, people quickly find out that they can make the BlenderBot say creepy or wrong things (or even criticize the bot’s nominal boss, mark zuckerberg), but that’s the purpose of publishing these demos.
As the head of research engineering at Facebook AI Research (FAIR), Mary Williamson, told The Verge earlier this month, many companies do not like to test their chatbot because what they say will be detrimental to business, as with tay of microsoft.
But for many researchers, the best way to improve these very bots is to launch them into the public arena, where they will be intensely challenged and manipulated in the multitude of unheard-of ways an unbiased engineer could dream of.
“This lack of tolerance for bots saying useless things, broadly speaking, is unfortunate,” Williamson said. “And what we’re trying to do is publish this very responsibly and push research forward.”
It’s interesting to compare Google and Meta in this regard, as Meta has certainly applied fewer restrictions to interacting with BlenderBot.
Craiyon, a website to create images using words
To use Craiyon, you simply have to write one, two or more words in the corresponding field. Right-clicking or pressing the key enterthe technology behind this app, will begin to create illustrations from the words that have been written. The process is fast and you will see the results immediately.
A total of nine images that can be enlarged by clicking on them and saved by right clicking (on the desktop) or by holding the image with your finger (on smart phones Y tablets).
There’s even a button Screenshot to create an image containing all nine jobs to share in the social networks or save as a keepsake.
