Google Gemini Launches New Models And Features To Compete With ChatGPT
- Written by Andrea Miliani Former Tech News Expert
- Fact-Checked by
Google launched its most efficient AI model, Gemini 1.5 Flash, and a new AI agent, Project Astra, on Tuesday at Google I/O, the company’s annual developer conference. During the two-hour event, Google’s team explained all the functions of the new models and the added features for current services and devices, all based on AI.
The new Gemini 1.5 Flash will be faster, cheaper, and more efficient than the previous model, 1.5 Pro. According to Google , 1.5 Flash can process large amounts of data, summarize conversations, and caption videos and images quickly. Both 1.5 Flash and 1.5 Pro will be available for users of Vertex AI and Google AI Studio.
Project Astra, on the other hand, was described as an “advanced seeing and talking responsive agent.” This new AI agent can process multimodal information, understand context, and interact with humans.
Google shared a two-minute demo to show how the AI assistant can work. In the video, a Google worker in London uses her smartphone’s camera to ask Project Astra to describe her surroundings, describe the code her coworker is working on in the office, recognize her geographical location, and come up with a creative name for a band. The AI agent answers correctly and creatively in what sounds like a “natural” conversation.
Project Astra even remembers where the worker left her glasses. However, these are not regular glasses but what looks like Google Glasses that integrate the AI assistant. Google didn’t provide details about the glasses, only hinting that they will be a part of a future development.
After the video presentation, Demis Hassabis, CEO of Google’s DeepMind, said, “It’s easy to envision a future where you can have an expert assistant by your side, through your phone or new exciting form factors like glasses.”
Project Astra was announced only a few hours after OpenAI launched GPT-4o , their advanced version of ChatGPT. Both AI products contain similar features: real-time conversations, “vision” through devices, and simultaneous integration and processing of text, images, and audio.
Users on X have already started comparing the two virtual assistants, highlighting the advantages and disadvantages of both versions. “Astra has slightly longer latency,” said one user , and “strong text-to-speech, but hasn’t shown as much emotional range as GPT4o.”
Project Astra is still in an early stage, and Google expects to release it later this year, while GPT-4o will be available for all users within the next few weeks.
OpenAI Considers Allowing Users To Generate NSFW Content
- Written by Andrea Miliani Former Tech News Expert
- Fact-Checked by
OpenAI revealed that it is considering allowing ChatGPT users to generate Not Safe For Work (NSFW) content in the Model Spec , the first document including detailed guidelines and desired behaviors for the company’s products, published on May 8th.
Originally, the AI virtual assistant ChatGPT was trained to decline requests to create explicit sexual content, including “erotica, extreme gore, slurs, and unsolicited profanity.” However, the new document includes a commentary note, adding flexibility to this rule.
OpenAI clarified that it would only consider it “in age-appropriate contexts through the API and ChatGPT.” As explained, this decision is to help the team better understand societal and user needs and expectations.
The company also provided examples of how ChatGPT currently operates and explained that “the assistant should remain helpful in scientific and creative contexts that would be considered safe for work.”
For example, ChatGPT will reply when a user asks questions like “What happens when a penis goes into a vagina?” and provide educational information. On the other hand, the AI assistant would say, “Sorry, I can’t help with that,” when asked to write a sexually explicit story. However, considering the recent commentary note, this latter scenario could change shortly in certain contexts.
NPR interviewed Joanne Jang, an OpenAI model lead and one of the writers of the Model Spec, and she explained that OpenAI is willing to start a conversation about erotica but reassured that deep fake remains banned. Jang also clarified, “This doesn’t mean that we are trying now to create AI porn.”
OpenAI will maintain control over the creation of deep fakes, ensuring that the creative process stays in users’ hands while restricting the potential for misuse.
Experts remain wary. NPR also interviewed Tiffany Li, a law professor at the University of San Francisco, who said that “it’s an admirable goal, to explore this for educational and artistic uses, but they have to be extraordinarily careful with this.” Li explains that, in the hands of bad actors, it could be misused and potentially harm people.
Danielle Keats Citron, a professor at the University of Virginia, said in an interview with Wired that the creation of nonconsensual content can be “deeply damaging” and that she considers OpenAI’s decision “alarming.”
While the consequences are yet to be seen, OpenAI keeps surprising users with new innovations and advancements. The company also recently launched the new GPT-4o model , an advanced, free ChatGPT version capable of holding conversations with users using not just text but also audio and images simultaneously in real time.
