To keep up with generative AI competitors like Anthropic and OpenAI, Google is rolling out updates to the free version of Gemini, its AI-powered chatbot. The updates aim to make the platform more powerful and more widely available.
Starting Thursday, Gemini 1.5 Flash, a lightweight multimodal model Google announced in May, will be available on the web and mobile in 40 languages and about 230 countries. Google says Gemini 1.5 Flash brings improvements in quality and latency, with particularly notable improvements in image reasoning and understanding.
What would be a boon for Google is that it might also be cheaper to run in the background.
When Gemini 1.5 Flash was launched, Google emphasized that the model was a “distilled” and highly efficient version of Gemini 1.5 Pro, designed for what the company describes as “narrow” and “high-frequency” generative AI workloads. Given the overhead of implementing a chatbot platform like Gemini (see OpenAI’s ChatGPT billings), Google is no doubt eager to seize cost-cutting opportunities, especially if those opportunities have the side effect of improving performance in other areas.
Beyond the new base model, Google says it’s expanding Gemini’s popup window to 32,000 tokens, which equates to about 24,000 words (or 48 pages of text).
Flash Gemini 1.5
Image credits: Google
Context, or context window, refers to the input data (e.g., text) that a model considers before generating output (e.g., additional text). Advantages of models with larger contexts include that they can summarize and reason about longer text snippets and files (at least in theory), and that in a chatbot context, they are less likely to forget topics that have been discussed recently.
The ability to upload files to Gemini for analysis previously required Gemini Advanced, the paid edition of Gemini that was built on top of Google’s $20-per-month Google One AI Premium plan. But Google says it will soon enable file uploads from Google Drive and local devices for all Gemini users.
“For example, you’ll be able to upload your economics study guide and have Gemini create practice questions,” Amar Subramanya, Google’s vice president of engineering, wrote in a blog post shared with TechCrunch. “Gemini will also soon be able to analyze data files for you, allowing you to uncover insights and visualize them using graphs and charts.”
In an attempt to combat hallucinations (where a generative AI model like Gemini 1.5 Flash makes things up), Google is introducing a feature that displays links to related web content below some Gemini-generated answers. English-speaking Gemini users in some territories will see a “bullet” icon at the end of a Gemini-generated paragraph with a link to websites (or emails, if you’ve allowed Gemini to access your Gmail inbox) where you can dig deeper.
The move follows revelations that Google’s generative AI models have a tendency to have very bad hallucinations, such as suggesting nontoxic glue in a pizza recipe and making up fake book reviews attributed to real people. Earlier this year, Google launched a “double-check” feature in Gemini designed to highlight claims from Gemini that other online sources either corroborate or contradict. But the links to related content appear to be an effort to make it more transparent about the sources of information Gemini might be relying on.
The question this reporter is asking is how often and how accurately Gemini will surface related links. TBD.
But Google is not waiting to flood the channels.
After launching Gemini in Messages on select devices earlier this year, Google is rolling out the feature to the European Economic Area (EEA), the UK, and Switzerland, with the ability to chat in new languages including French, Polish, and Spanish. Users can access Gemini in Messages by tapping the “Start Chat” button and selecting Gemini as their chat partner.
Google is also launching the Gemini mobile app in more countries and expanding access to Gemini to teens worldwide.
In June, the company launched a Gemini experience for teens, allowing students to sign up using their school accounts, but not in all countries. Over the coming week, that will change, as Gemini will be available to teens in all countries and regions where Gemini is normally available to adults.
Alongside the rollout, Google says it’s putting in place “additional policies and safeguards” to protect teens, without going into specifics. There’s also a new teen-friendly onboarding process planned, as well as an “AI learning guide” to, as Google puts it, “help teens use AI responsibly.”
There’s a raging debate over whether kids are using generative AI tools as intended or abusing them. Google is certainly keen to avoid headlines suggesting that Gemini is a plagiarizing essay generator or capable of giving teens ill-conceived advice on personal issues, and therefore taking steps to prevent the worst from happening.
.jpg)
.webp)
0 Comments