Sunday Rundown #101: Talking Videos & Holiday Hallucinations

💰Sunday Bonus #61: Comparison page for 23 text-to-image models.

Daniel Nest

Jun 08, 2025

∙ Paid

Happy Sunday, friends!

Welcome back to the weekly look at generative AI that covers the following:

Sunday Rundown (free): this week’s AI news + a fun AI fail.
Sunday Bonus (paid): an exclusive segment for my paid subscribers.

Every Sunday Bonus in one place

Let’s get to it.

🗞️ AI news

Here are this week’s AI developments.

👩‍💻 AI releases

New stuff you can try right now:

Anysphere released Cursor 1.0, which can now help review your code, recall past errors, and multitask across dozens of background jobs.
ElevenLabs dropped v3 (alpha), the most expressive TTS model, supporting 70+ languages, emotion tags, multi-speaker dialogue, and more.
Genspark launched AI Secretary, an all-in-one assistant for Gmail, Calendar, Drive, and Notion that auto-replies, schedules, summarizes, and more.
Google news:
1. NotebookLM finally lets you publicly share notebooks so others can view and interact with your notes and sources.
2. Gemini 2.5 Pro got another upgrade, improving its #1 lead on the Chatbot arena by 30 ELO points. (Free to try in Google AI Studio.)
3. Google AI Pro and Ultra users can now use scheduled actions in the Gemini app to schedule tasks.
HeyGen upgraded Avatar IV with prompt-controlled gestures and hyper-real micro-expressions at 1080p video. (I first wrote about Avatar IV here.)
Higgsfield AI launched Speak, a feature that turns scripts into motion-driven talking videos, available to Pro and Ultimate plans.
HubSpot released a deep Research connector for ChatGPT that lets users pull their CRM insights directly into chats for further analysis.
Luma Labs launched Modify Video, letting you change the style, objects, settings, and more in recorded videos.
Manus introduced Manus video generation, which can turn a single prompt into a fully storyboarded, animated video. (In early access for paying accounts.)
Microsoft news:
1. Bing Video Creator generates 5-second videos from text prompts, powered by OpenAI’s Sora model. (Free for everyone on mobile.)
2. Copilot Shopping is a smart assistant that finds better deals and curates shopping picks tailored to your needs.
3. Photos app now comes with an AI-powered “relight” feature that can adjust lighting and an AI search that helps find photos by topic and theme.
Mistral released Mistral Code, a full-stack AI coding assistant that bundles models, in-IDE tools, and local deployment. (Request access.)
OpenAI news:
1. A lightweight version of “Memory” is rolling out to free ChatGPT users, letting ChatGPT tailor responses based on recent conversations.
2. Connectors let ChatGPT incorporate live context from Outlook, Drive, Gmail, and more into ongoing chats.
OpenAudio launched S1, a state-of-the-art TTS model with actor-level expressiveness, emotion tags, and top-tier accuracy.
Stanford’s Scaling Intelligence Lab released Tokasaurus, a really fast LLM inference engine that runs AI models up to 3x quicker.
Suno upgraded its Song Editor with track editing, stem splitting, full-song reference uploads, and creative sliders.
TikTok introduced AI-powered Smart Keyword Filters to help users filter out unwanted content and topics.
Topaz Labs has an upscaler called “Bloom” that uses AI to make images 8X bigger while adding creative detail.

🔬 AI research

Cool stuff you might get to try one day:

Alibaba’s QwenLong-L1 reinforcement learning framework boosts LLMs from short-context to long-context reasoning.
Apple is expected to unveil Apple Intelligence updates at WWDC, like live translation via AirPods and AI tools for developers.