O Google ran to throw the bard and react to all the buzz around the ChatGPT. But it looks like he wasn’t ready. Humans need to train the chatbot to give the right answers. The problem is that sometimes even these humans don’t know what the right answers are. Then, they do what we do in some school exams: “kick”.
The information is from a report by the Business Insider. The publication spoke with four employees hired by Appen, a company that provides services to Google. They spoke on condition of anonymity.
Appen employees are “raters”, or “evaluators”, in free translation. Generally, they judge ad placement and the accuracy of search algorithms.
Since January, however, much of their work has been reviewing the responses Bard gives users to ensure they are up to date and accurate, per Google standards.
Do not know? “Kick”
The work happens as follows: the system displays the user’s prompt (a request, a question, a task, among other possibilities) and two responses suggested by the Bard.
The evaluator chooses the best of the two and writes a justification. Artificial intelligence itself uses this justification to improve its capabilities.
For each rating, raters have between 60 seconds and several minutes.
The problem is that, sometimes, not even the employee can say what the right answer is, especially if it is a subject that he does not master.
Only the workers receive by evaluation. So taking too long to find the right answer means less money in the account.
What do they do then? Simple: they “guess” the answer that seems most correct and move on to the next one.
One of the officials interviewed by the Business Insider said it’s frustrating not having enough time to research and figure out which option is right.
“(We need) three hours of research to complete a 60-second task,” he laments. “It’s a good way to explain the problem we have to deal with at the moment.”
Bard (and other AIs) have accuracy issues
Since ChatGPT was launched, it has been speculated that it may one day replace traditional search engines. Google then went into a state of alert.
In February, the company announced its chatbot: Bard. What was meant to be a sign that the company was in the running, however, turned into a huge headache.
The problems started right after the presentation. One of the answers provided by the Bard was wrong. Investors reacted badly, and Google lost $100 billion of market value.
Justice be done: ChatGPT makes mistakes too, and it makes bad mistakes at times.
Google, then, started to ask its own employees to take some time out of the office to write answers for the robot, according to subjects that they dominated.
Internally, this whole sequence of events caused irritation, as the workers considered that Bard was not ready and does not live up to Google’s quality standards.
The company had to step back and say that it didn’t see Bard as a search replacement, but as a complement. Even so, Sundar Pichai, CEO of Google, ensures that its AI is the best on the market.
By all indications, Bard and his employees have a lot of work ahead of them.
With information: Business Insider