AI News Flash | OpenAI's New "Codex" Tool Built Into ChatGPT

Hello from the world of AI! 🚀In today's newsletter, we cover the following hot topics.　・Google to start showing ads in its new search feature "AI Mode"　・Alibaba unveils open-source AI "Wan2.1-VACE" — a step toward unifying video production　・Meta launches "Llama for Startups," an AI support program for young companies

GOOGLE

Google to start showing ads in its new search feature "AI Mode"

Google has revealed plans to trial ads within"AI Mode," its AI-driven search experience.

Source: TechCrunch

Ads will appear naturally below or within the AI's answer, depending on intent. Unlike traditional search, AI Mode lets people dig deeper through a conversational format.Google says the goal is to surface useful ads for action-oriented queries — like "how to build a website" — helping users start a project, then keep refining their idea, content strategy, or target audience through conversation.Given that most of Google's revenue comes from ads, bringing them into AI Mode is a natural step. In Q1 2025 the company loggedroughly $66.8 billion in ad revenue alone. Still, users aren't fully sold: a U.S. survey (CivicScience) found36% of adults feel uneasy about AI-powered ads and say it makes themless likely to buy from those brands.Google says advertisers running"Performance Max," "Shopping," or "Search" campaigns withbroad match settings will be eligible. Rollout starts with U.S. users, on both search andshopping-style ad formats across desktop and mobile.

Source: TechCrunch

Google expands ads in its "AI Overviews" search-summary feature

In a Wednesday update, Google outlined plans towiden ad coverage within "AI Overviews," which has AI summarize search results — starting with search/shopping ads for U.S. desktop, then expanding todesktop and mobile in other English-speaking markets.Ads within AI Overviews are inserted into theAI-generated summary text and appear only when relevant to the query and answer, carrying an explicit"Sponsored" label.Google had already been trialing similar placements on U.S. mobile since October 2023 (10/2023).

Publisher concerns, and Google's response

Some publishers worry this couldhurt their own ad revenue. Google says it's taking that feedback seriously while developing AI search and ad products.

A broader industry trend

Folding ads into AI products isn't unique to Google.　・AI search engine"Perplexity" has shown ads since last November (11/2024).　・Microsoft trialed ads in Copilot a few years back.　・OpenAI has hinted it may add ads to complement paid subscriptions.

ALIBABA

Alibaba unveils open-source AI "Wan2.1-VACE" — a step toward unifying video production

Alibaba has unveiled "Wan2.1-VACE," an AI model that could reshape video production. Open source, it covers everything from generating video to editing it.VACE is part of Alibaba's broader "Wan2.1" video platform, positioned asthe industry's first open model covering diverse video needs in a single solution.Work that once required stitching together multiple tools can now happen in one interface via VACE — a shift that could reshape content production.So what can it actually do? It generates video from text prompts, still images, or short clips.Its editing tools are strong too: referencing images or frames to guide edits, refreshing footage via "repainting," partial fixes, and extending duration. Alibaba says this flexibility unlocks users' creativity.

Source: www.alibabagroup.com

To turn a few still images into characters interacting on video, VACE is ideal. This open-source AI adds natural motion to photos, bringing them to life."Repainting" allows fine control — transferring one character's motion to another, adjusting movement flow, depth, or color.It also supports adding, fixing, or removing specific areas without affecting the surroundings — great for delicate edits, and a step apart from older tools that couldn't touch one part without disturbing the background.VACE can also extend a frame and naturally fill the new space, making video richer and more dynamic.You can sketch a motion path onto a still image to design how an object moves, swap characters or objects using a reference image, or animate a character with precise pose control.Alibaba also says VACE can turn a portrait-oriented image into a wide, landscape video, automatically filling in natural detail from other images or input data — a big potential efficiency gain for video production.VACE is more than a video generator, though. Several technologies power it, including the "Video Condition Unit (VCU)," a core module that consistently processes text, images, video, and masks.It also has a "Context Adapter" that lets the AI grasp timing and spatial structure, enabling more refined, natural video output.Alibaba expects use across social video, promotional footage, film editing, and educational content.

Alibaba releases "Wan2.1-VACE" in two versions to fit different needs

Building an AI model takes massive compute and huge training data, making it costly. Against that backdrop, open-sourcing "Wan2.1-VACE" marks a major step for the industry.Alibaba says that as more people can create high-quality visuals quickly and cheaply, the barrier to using AI drops — with small businesses and individual creators standing to benefit most.In short, Alibaba wants to make powerful AI tools easy to access for a wide range of people. Democratizing strong technology is always welcome.There isn't just one version, either: a large model with1.4 billion parameters for high-performance systems, and a lightweight130 million-parameter version for constrained environments. Both are free to download via Hugging Face, GitHub, or Alibaba Cloud's "ModelScope."

META

Meta launches "Llama for Startups," an AI support program for young companies

To help drive generative AI forward, Meta has launched"Llama for Startups," hoping to speed adoption of its "Llama" model. It offershands-on support from Meta's technical team plus funding for qualifying companies: those incorporated in the U.S., with under $10 million raised, at least one developer, and an active generative AI app. Applications closeMay 30.Selected companies getup to $6,000 a month for up to 6 months to help build and refine their AI, with Meta specialists helping explore advanced Llama use cases.It's part of a push to cement Meta's lead inopen-source AI.Llama has already passed 1 billion downloads, yet still faces fierce competition from Alibaba's Qwen, Google, and DeepSeek.Llama has also hit some recent snags, and this program is seen partly as a response.Meta's in-development "Llama 4 Behemoth" has been delayed — the Wall Street Journal reports it's fallen short internally on some benchmarks. In April, allegations also surfaced that Meta used an unreleased version of"Llama 4 Maverick" to post a high score on benchmark site"LM Arena," since the released model didn't match what was ranked — fueling criticism of Meta's transparency and raising questions about its credibility in a fast-moving industry.

Metastill has big ambitions for its Llama-based generative AI products. Last year it projected$2–3 billion in 2025 generative AI revenue, andup to $1.4 trillion by 2035.Meta hasrevenue-sharing deals with several companies hosting Llama, and unveiled anAPI for customizing Llama too. Its assistant"Meta AI" may go paid down the road, with ads and extra features, perCEO Mark Zuckerberg on the Q1 earnings call.This doesn't come cheap, though.Meta'sgenerative AI budget for 2024 had alreadytopped $900 million, on track to pass $1 billion by year-end. Infrastructure costs run high too — Meta plans$60–80 billion in 2025 capex, mostly for new data centers.

In Closing

We hope today's content helps you make the most of AI technology and stay informed.If you found this newsletter useful, please feel free to share it with colleagues or friends who might be interested.We will keep curating and delivering the most notable topics so that we can continue bringing you valuable information.Look forward to seeing you in our next newsletter.Thank you for your continued support.