🥴Character.AI Makes Deepfakes Fun

Today we will discuss:

• Sponsored: LangWatch—Smarter LLM monitoring, testing, and optimization

• Character.AI unveils AvatarFX—New model lets users animate AI characters and real images, raising deepfake concerns and creative possibilities

• Don’t miss this week’s Workflow Wednesday—AI learning that sticks, sharpen your prompting skills with LearnPrompting.org, use Jenni AI as your research + writing assistant, and automate your AI-powered reading list with Notion, Zapier, and ChatGPT; now available at only $20/month! 

• Grok upgraded—xAI adds vision, multilingual speech, and real-time voice search to its chatbot

• In other AI news today—OpenAI exec says the company would buy Google’s Chrome browser if given the chance, Microsoft 365 Copilot gets a redesign with new search and notebook features, Nvidia’s AI assistant adds plugins for Spotify and Twitch, and Apple removes the 'available now' claim from its Apple Intelligence page

• Saved the best for last—10 must-try AI tools

 OpenSource Monitoring, Evaluating & Optimization for your LLM-apps

Today, LLM testing is often scattered across notebooks, dashboards, and scripts, making it slow, unreliable, and difficult to scale effectively.

We’ve heard the same pain points from teams struggling with LLM performance testing:

• "I don't have a dataset to evaluate yet"

• "How can we measure the quality when there is no defined golden answer in our case?"

• "I have no idea how we should get started with evals, should we use LLM-as-a-judge?"

LangWatch brings you a smarter, and more reliable way to test and optimize your LLMs. 

Here’s what you’ll love about the newest release of LangWatch: The Evaluations Wizard:

• Build your own Evaluators: Use 50+ built-ins, LLM-as-a-judge, your own, or a human-in-the-loop review.

• Evaluate full pipelines: Chain LLM calls together and test across realistic, multi-step workflows.

• Launch in minutes: Upload test data, generate synthetic examples, or connect your pipeline—no setup headaches.

• DSPy optimizers built-in: Optimize prompts and model configs programmatically, directly in LangWatch.

• CI/CD Ready: Run batch tests via SDKs or CLI and get detailed reports, costs, and latency fast.

Whether you're an AI developer, AI PM, domain expert or any other person involved in building LLM-apps this is your new command center for LLM quality. 

This is a big change in how teams will be Evaluating the performance of their LLM-apps.

🔍 Try the LangWatch for free now
📘 Read the launch blog post

Join our Discord
Fully Open-source: Check-out our Github Repo

Character.AI’s new AvatarFX animates AI characters and real people’s photos in multiple styles, voices

Key Points 

• AvatarFX lets users animate AI characters and real images, offering more than just basic text-to-video generation.

• Character.AI says it will watermark videos, block animations of minors, and filter celebrity images to prevent misuse.

🤖News - Character.AI, the platform known for letting users chat and roleplay with AI-generated characters, has announced a new tool called AvatarFX. Now available in closed beta, AvatarFX can animate characters in a range of styles and voices, from realistic human figures to 2D animal cartoons. 

🤔What's the most interesting point here? Unlike OpenAI’s Sora, AvatarFX is not limited to just text-to-video. It also lets users animate existing images, including photos of real people.

😵‍💫Why this is problematic - While the technology opens up creative possibilities, it also raises some serious concerns. People could upload photos of celebrities or individuals they know and create videos where they appear to do or say things they never actually did. 

Although deepfakes have been around for some time, building this kind of tool into a widely used platform like Character.AI could make it easier for the technology to be misused.

🙄Safeguards are promised, but questions remain - To address these risks, Character.AI says it will watermark all AvatarFX videos to show they are AI-generated. The company also plans to filter and modify images of real people, block the animation of minors, and restrict the generation of videos featuring major celebrities and politicians. 

But since AvatarFX is still in limited testing, it remains to be seen how well these safeguards will work in practice.

Read the full story here. 

We just dropped the 16th edition of Workflow Wednesday—built for AI-minded professionals who want to do more than just read about AI.

This week’s theme: AI Learning That Sticks

Sharpen your prompting skills with LearnPrompting.org, turn Jenni AI into your personal research + writing assistant, and set up an automated AI-powered reading list with Notion, Zapier, and ChatGPT.

Join here

xAI gives Grok vision, multilingual speech, and real-time voice search features

Key Points 

• Grok Vision lets users point their phone at objects and ask questions, similar to other real-time vision tools.

• xAI also launched multilingual audio support and real-time search in Grok’s voice mode, expanding user capabilities.

• Android users must subscribe to xAI’s $30/month SuperGrok plan to access the latest voice features.

☕News - xAI launched Grok Vision, a feature that lets users point their phone’s camera at objects like products, signs, and documents, then ask questions about them. This is similar to the real-time vision features already available in tools like Google’s Gemini and ChatGPT. 

For now, Grok Vision is available on the iOS version of the Grok app, but Android users will have to wait a bit longer. 

👾What’s more? In addition to Grok Vision, xAI also introduced some other notable updates. Grok now supports multilingual audio, allowing users to interact in several different languages. There’s also a new real-time search feature in Grok’s voice mode, which enables users to search for information quickly while speaking.

However, to access these new voice features, Android users must be subscribed to xAI’s $30-per-month SuperGrok plan.

Read the full story here.