And now, we’re hearing more on this front including how new updates would fix most of the errors and enhance its ability to listen and follow instructions as well as provide the right replies. But with the good news comes some bad, and that includes how the company is yet to focus on Gemini.
Gemini has been the center of serious controversy in the past few weeks after it failed to produce the right depictions of different individuals. Moreover, the latest versions of Geem are said to be doing the exact opposite, as confirmed by DeepMind’s top researcher.
The tech giant also shed light on how it has worked toward fixing specific bugs that arise if and when the user tried to switch a topic while communicating with Gemma and limiting occurrences with terms such as Sure that were commonly seen at the start of different sentences.
Overusing specific terms including those seen as replies for queries is quite the common practice and therefore serious error for many generative AI models.
It’s become so much simpler for different journalists as well as researchers who keep on detecting AI usage across academic assignments. And before we forget, ChatGPT makes use of sentences such as ‘as of last knowledge update’ too.
The training data seen on Gemma failed to alter much with these updates as per the researchers but such reinforcement algorithms were altered as part of the new update. As mentioned by Amazon, it defines RL as machine learning of human trial and error. In this way, it puts out a reward and punishment form of action to conduct AI model training in all of its glory.
The new updates for Gemma mean a lot of its tendencies may now be altered with specific prompts. For example, requesting Gemma to carry out replies in paragraphs would limit the ability of AI to give rise to bullet point answers. So Gemma would now be working with goals that are linked to word count so the AI wouldn’t be going on forever.
What’s interesting is how Gemma was designed and customized by Google using the same tech that we’re seeing take center stage with Gemini. It was first publicly rolled out in February when developers were adding more to AI tools.
Today, Gemma is open but not exactly open-source. So this means it can be used for the sake of third-party use but with greater tech restrictions. Interestingly, Nvidia is giving rise to Gemma as an integral component of its Chat with RTX endeavor. Meanwhile, Opera just mentioned its goal linked to Gemma getting bigger updates.
Meanwhile, Gemini sees more restrictions as time passes on. This is regarding the various styles of content it has authorization to produce. In the past month, we saw Google stop Gemini from answering any prompts regarding controversial subjects like the elections or politics in general.
Let’s not forget the biggest hurdle that people have been dealing with for so long has to be linked to the AI model’s ability to produce pictures of individuals which continues to be barred. Remember, when the whole fiasco regarding image generation began, Google explained how it was more cautious on this front than what it had intended in the past. This gave rise to a plethora of inaccurate images which was an outcry of hundreds.
Image: DIW-aigen
Read next: New Find My Device Version Debuts in US and Canada, Global Expansion Planned
