This past week brought several interesting developments, from tools you can start using today to glimpses of what’s coming next. We’ve broken down some of the highlights based on how easy they are for the average person to jump into.
Easy to Use: Get Started Right Away!
These updates bring new capabilities to tools you might already use or are easily accessible:
- Deep Search for Everyone on ChatGPT: OpenAI has made its “deep research” feature available on the free plan of ChatGPT. Previously for paid users, this feature does a deep dive search on the internet to pull out a lot of information on a topic. Free users get five deep searches per month. Paid users (Plus, Team, Pro) get access to a “lightweight version” to increase rate limits, although the exact details for paid users are a bit unclear.
- Perplexity Assistant for iOS: The Perplexity app on iOS has a new Assistant that aims to be like a better version of Siri. Besides answering questions, it can help you with tasks like playing media, drafting emails, moving meetings, booking rides, making reservations, and setting reminders. While some features are still being refined, examples show it successfully opening booking pages or listing calendar events.
- Microsoft Recall: A new feature from Microsoft that’s like a browser history for your entire computer. It lets you go back and see anything you were working on in any app, almost like rewinding time. You can even use AI to search your past activity. Recall is rolling out as an opt-in feature, meaning it’s not on by default. Your data is processed and saved locally on your device, not sent to Microsoft’s cloud.
- Better Microsoft Search and “Click to Do”: Microsoft is improving its search using AI to better understand what you’re looking for, even if you don’t know the exact file name. They’re also rolling out “click to do,” which lets you summarize, rewrite, or copy text and images directly from your screen.
- Grock Chatbot Can Now “See”: XAI’s Grock chatbot now has vision capabilities, similar to models from Gemini and OpenAI. This means you can show it a picture (like through the mobile app’s camera) and chat with it about what it sees. Examples show it describing a workstation or a picture on a wall.
- Ray-Ban Meta Glasses Get Live Translation: If you have Ray-Ban Meta smart glasses, a new feature allows for live translation. Someone speaking a different language to you will be translated into your preferred language right into the headphones on the glasses. You can also download language packs for offline translation.
- YouTube Testing AI Video Overviews: YouTube is experimenting with an AI overview feature for search results. Instead of a text summary, it might show clips from videos that are most relevant to your search, letting you find information quickly without necessarily clicking into a full video. This is currently being tested with a small group of YouTube Premium users.
- Character AI Adds Visuals to Chats: Character AI, a popular platform for chatting with fictional characters, is rolling out avatar effects that generate visuals for the characters you talk to. This makes chatting feel more interactive and less like just talking to a text bot. This feature is rolling out now, with early access available.
Moderate Use: Worth Exploring for Specific Tasks!
These tools might require using a specific platform or have a slightly steeper learning curve, but offer powerful creative abilities:
Descript Testing AI Video Editing: The video editing tool Descript is testing new AI features that allow you to edit videos by simply chatting with a bot. You can ask it to write a script draft, cut down a video, or add visuals like chapter titles or stock overlays. This feature is not yet publicly available.
Microsoft 365 Copilot Gets Smarter: Updates to Microsoft 365 Copilot add more “agentic” features. This includes AI search, a new creation experience, notebooks for organization, and an “agent store”. These agents can be optimized for specific tasks like research or data analysis, and can connect to other tools like monday.com or Dropbox. Access is expected for most users in late May.
Generate Videos Affordably with LTX Studio: The LTX Studio platform has added Google’s V2 video generation model. This makes LTX Studio potentially the least expensive way to generate videos using V2, costing significantly less per second than using Google’s own cloud platform. You can generate videos from images or text prompts.
Adobe Firefly Updates: Adobe released new versions of its Firefly AI image generation models. Users can now choose between different models within the Firefly web app, including the new Firefly Image 4 and Ultra versions.
Crea AI for Image and 3D Editing: The company Crea AI rolled out features to edit images directly in chat using the ChatGPT image model. They also introduced a feature called “Stage” which lets you create 3D environments from images or text and even manipulate objects within the scene.
Advanced: For the Tech Enthusiasts and Developers
These updates are primarily aimed at those with more technical skills or specific business needs:
- OpenAI Planning an Open Model: Rumor has it OpenAI is planning to release a new AI model around June that will be free to download and run on your own computer. It aims to perform better than other open models and might even be able to call upon OpenAI’s larger models via API for complex tasks, though this is speculative. This is exciting for those who want to run models locally.
- New APIs Released: OpenAI’s image generation technology (like used in ChatGPT) is now available via an API for developers. XAI also released an API for their Grock 3 Mini model, which shows strong benchmark performance at a lower price compared to some other models.
- AI Avatars Holding Products: A company called Argil is rolling out a feature where AI-generated avatars can hold up actual products. This has big implications for e-commerce and branding, allowing companies to create spokespeople for their products.
- Improved AI Lip-Syncing: A company called Tavis released a new model claimed to be the best available for lip-syncing. While the technology is still improving and can look a little unnatural (“uncanny”) to some, it’s getting much closer to realistic.
Important News and Insights
Beyond new tools, there were also significant discussions and partnerships:
- Washington Post Partners with OpenAI: The Washington Post is partnering with OpenAI, allowing OpenAI’s search features to access Washington Post content. This kind of partnership seems to be a strategy for AI companies to avoid potential lawsuits related to using news content.
- Anthropic’s Focus on AI Safety: AI company Anthropic released several pieces emphasizing their focus on understanding and addressing the potential harms of AI beyond just “doomsday scenarios”. They discussed the need to look at physical, psychological, economic, societal, and individual impacts. They also shared studies on detecting malicious uses of their Claude model (like for political bot farms or coding malware) and stressed the urgency of understanding how these models “think” (“interpretability”). This focus seems to influence their development speed compared to some other companies.
- The Oscars’ View on AI: The Academy governing the Oscars announced that using generative AI or other digital tools in filmmaking will neither help nor hurt a film’s chances of nomination. They will still consider the degree to which human creativity was at the heart of the work.
- AI Companies Eye Google Chrome: In a recent court case, both OpenAI and Perplexity expressed interest in bidding to buy Google’s Chrome browser if it were to become available. This suggests potential plans for building AI-first browsers.
- DeepMind CEO on AI Consciousness: Demis Hassabis, CEO of Google DeepMind, discussed the possibility of AI systems acquiring something similar to self-awareness in the future, noting that understanding “self and other” could be the beginning of this.
That wraps up some of the key AI updates from this past week! As you can see, AI is moving quickly, bringing new capabilities to users, tools for creators, and sparking important discussions about its future.