Kaopiz Tech Briefs: SearchGPT; Google Gemini; Meta AI; Adobe AI; Cybersecurity
1. OpenAI announces SearchGPT, its AI-powered search engine:
OpenAI has introduced SearchGPT, a new search feature designed to compete with Google and Bing.
Unveiled recently, SearchGPT offers timely answers by drawing from web sources. The interface is similar to OpenAI's ChatGPT, where users type queries and receive information, photos, and links from the web. Users can ask follow-up questions and explore related searches in a sidebar.
SearchGPT can tailor results based on the user's location, collecting and sharing general location information with third-party search providers to enhance accuracy. Precise location information can be shared via a toggle in the settings menu.
Powered by OpenAI models GPT-3.5, GPT-4, and GPT-4o, SearchGPT is currently available as a prototype to a small group of users and publishers, with a waitlist for broader access. There are plans to integrate some features of SearchGPT into ChatGPT in the future.
OpenAI believes that by enhancing the conversational capabilities of their models with real-time web information, finding relevant information can be faster and easier. SearchGPT prominently cites and links to publishers with clear attribution, and OpenAI is collaborating with publishers to design the user experience. Website owners are given control over how their content appears in search results.
SearchGPT is separate from OpenAI’s generative AI foundation model training, ensuring that sites can be included in search results even if they opt out of generative AI training. OpenAI emphasizes its commitment to supporting a thriving ecosystem of publishers and creators.
2. Google makes its Gemini chatbot faster and more widely available:
To keep pace with rivals like Anthropic and OpenAI, Google is updating its free-tier Gemini AI chatbot.
Starting Thursday, the Gemini 1.5 Flash model will be available on the web and mobile in 40 languages across approximately 230 countries. This new model promises improved performance with better quality, faster response times, and enhanced reasoning and image understanding.
The Gemini 1.5 Flash model is described as a “distilled” version of Gemini 1.5 Pro, designed to handle specific, high-frequency generative AI tasks more efficiently. This optimization may also reduce operational costs for Google, an important factor given the expenses associated with running large-scale AI platforms. In addition, Gemini’s context window is being expanded to 32,000 tokens, equivalent to about 24,000 words or 48 pages of text.
This upgrade allows the chatbot to process more content at once and respond more quickly. Amar Subramanya, VP of Engineering for Gemini, highlights that these updates will result in “across-the-board improvements in quality and latency,” with notable advancements in reasoning and image processing capabilities.
3. Meta AI gets new ‘Imagine me’ selfie feature:
Meta AI, Meta’s AI-powered assistant for Facebook, Instagram, Messenger, and the web, has expanded its capabilities to support more languages and create stylized selfies.
Starting today, users can direct their questions to Meta’s latest AI model, Llama 3.1 405B, which is designed to handle more complex queries than its predecessor. The new model, Imagine Yourself, generates images based on user photos and prompts like “Imagine me surfing” or “Imagine me on a beach vacation.”
Currently in beta, it can be accessed in Meta AI by typing “Imagine me” followed by any non-NSFW request. Meta has not revealed the specific data used for training Imagine Yourself, but its terms of use indicate that public posts and images on its platforms may be utilized. This policy, along with the complex opt-out process, has faced criticism from some users. In addition to Imagine Yourself, new editing tools are being introduced in Meta AI, allowing users to add, remove, or alter objects with prompts such as “Change the cat to a corgi.”
领英推荐
Next month, an “Edit with AI” button will be added for finer adjustments, and new shortcuts will soon be available for sharing AI-generated images across Meta apps.
4. Adobe releases new Firefly AI tools for Illustrator and Photoshop:
New Firefly tools for Photoshop and Illustrator were released by Adobe on Tuesday, providing graphic designers with enhanced options for utilizing the company’s AI models.
These new features enable creative professionals to input brief prompts and receive AI-generated textures or images, which could otherwise take hours to produce. A new tool, “Generative Shape Fill,” is being introduced in beta for Illustrator. This feature allows users to add details and textures to shapes via text prompts or by selecting a style reference. It is powered by an updated beta version of Adobe’s Firefly Vector model.
For Photoshop, Adobe’s Firefly text-to-image generator is being made widely available. Users can now create AI images directly within the application by pressing “Generate Image” and providing a description. This functionality, powered by Adobe Firefly’s Image 3 Foundation model, was previously accessible only in beta.
Despite not charging a premium subscription for generative AI like some competitors, Adobe is still experiencing positive returns from these innovations.
5. Hidden Threats in GitHub Repositories: Malware Targets Developers:
A new malware campaign is spreading through GitHub, posing a significant threat to developers. Disguised as legitimate code, the malware strains, named Stargazer and Goblin, have been discovered infiltrating open-source repositories.
These malicious codes are cleverly hidden in repositories that appear harmless, leading unsuspecting developers to download and integrate them into their projects. Once installed, the malware can exfiltrate sensitive information and establish backdoors for further exploitation.
Its stealthy nature makes it difficult to detect, as it blends seamlessly with legitimate code. The campaign has raised concerns about the security of open-source platforms and the potential for widespread damage.
Developers are urged to exercise caution when sourcing code from GitHub and to implement strong security measures to protect against such threats. Regular code reviews, using trusted sources, and employing advanced security tools are recommended to mitigate the risk of malware infections.
As cybercriminals continue to evolve their tactics, vigilance in the software development community is more crucial than ever.
?? Explore the possibilities with Kaopiz – your trusted IT outsourcing partner.
?? Click here to visit our website and learn about out services! ?? #Kaopiz #ITOutsourcing #AI #ComputerVision # GenerativeAI # AIChatbot #CloudComputing #OpenAI # Microsoft #Amazon #Apple