Google / Gemini

AI at Google and DeepMind: the Gemini family, Veo video models, research and the product ecosystem.

409 picksRelated topicsOpenAI / ChatGPT Anthropic / Claude Multimodal

Latest picks

241–260 of 409

May 21Thursday

The Verge · AIMay 21, 00:41

You can now remix other people’s YouTube Shorts with AI

Google added a “reimagine” option to YouTube Shorts Remix, letting users use Gemini Omni to restyle clips, alter contents, or insert themselves into other people’s videos, while creators can enable or disable reimagining for their uploads.

Why it matters: HKR-H/K/R all pass: Google is adding Gemini Omni to the Shorts remix flow with a named reimagine mode and creator controls. It is a meaningful platform feature, not a model release, so it sits in low featured.

The Verge · AIMay 21, 00:00

Google Search’s AI Evolution Includes More Ads

Google is adding Gemini-generated product explanations to Search ads for product queries, including Sponsored Product placements and some ads with built-in chatbots; the snippet cites a compact espresso pod machine example, but the post does not disclose rollout scope or pricing.

Why it matters: HKR-H/K/R all pass: Google is inserting Sponsored Product units and ad chatbots into Gemini shopping results. Scope, pricing, and performance data are not disclosed, so this stays in low featured rather than a major product-release band.

Take

Google is spending Search’s remaining trust on ad inventory, and the placement is ugly. In product queries, Gemini will generate a purchase explainer for a Sponsored Product; Google’s own example is a “compact espresso pod machine” search showing a Nespresso Vertuo Up. Some ads also get built-in chatbots. Rollout scope and pricing are not disclosed, so this reads as a product-direction signal rather than a complete commercial launch.

I don’t buy the “better shopping help” framing. Classic Search ads were constrained by labels, bidding, and ranking slots. AI Overviews already blur answers and source material on the same screen. Now Gemini is writing the reason to buy the sponsored item. That moves model output from retrieval into sales copy. Great for advertisers; another withdrawal from Search credibility.

May 20Wednesday

The Verge · AIMay 20, 22:12

It’s Make-or-Break Time for AI Labeling Systems

Google announced at I/O an expanded ability to verify SynthID markers on AI-generated images, while C2PA Content Credentials also targets origin metadata for image, video, and audio files; the RSS snippet does not disclose the full rollout scope or verification limits.

Why it matters: HKR-H/K/R all pass, but the post lacks full coverage scope, rollout terms, and adoption data. This is a mid-weight Google/C2PA provenance update, not a must-write release.

The Verge · AIMay 20, 21:24

If Google Can’t Make AI Agents Useful, Maybe No One Can

The Verge says Google announced multiple AI agents at I/O 2026 for information gathering, event planning, and inbox or calendar summarization; the RSS snippet says the agents can run continuously in the background, but the post does not disclose launch timing, pricing, or evaluation results.

Why it matters: HKR-H and HKR-R are strong because the Verge frames Google agents as a sector test; HKR-K is limited to background-running agents. Missing launch timing, pricing, and evals keeps it in the lower featured band.

AI Chat-Group Daily (群聊日报)May 20, 14:51

2026-05-19 Chat Group Daily

The chat group daily says Karpathy joined Anthropic's pretraining team, and cites Stainless shutting down hosted services after acquisition plus Google I/O announcing Gemini 3.5 Flash and a $100 subscription tier.

Why it matters: HKR-H/K/R all pass, but this is a chat-daily roundup with secondhand claims and no disclosed primary links, appointment details, or product specs, so it lands at the lower featured band.

Synced · WeChatMay 20, 11:49

After I/O, Google turns the search box into an agent entry point

Google announced Gemini 3.5 Flash at I/O and added AI Mode directly to Search; the company said its AI services now process over 3.2 quadrillion tokens per month, with more than 8.5 million developers using Gemini.

Why it matters: HKR-H/K/R all pass: Google I/O combines a model update, Search distribution, and concrete usage numbers. AI Mode inside the search box is heavier than a routine feature release, so it clears the same-day must-write band.

Take

Google’s sharpest move is putting Gemini 3.5 Flash inside the Search box, turning search intent into agent dispatch. AI Mode already has over 1 billion monthly users, Gemini handles 3.2 quadrillion tokens per month, and 8.5 million developers use it. Those numbers matter more than the 76.2% on Terminal-Bench 2.1. Benchmarks get matched; default placement inside Search does not.

I don’t buy the “Gemini 3.5 era” framing yet. Google shipped Flash, while Gemini 3.5 Pro is still internal, with Pichai promising “next month.” This smells like an entry-point land grab: wire agent behavior into 13 billion-user products first, then let the flagship model arrive later. For builders, Antigravity and Spark inside Workspace are the actual surface area. The singularity quote is conference theater.

Latent SpaceMay 20, 11:34

Google I/O 2026: Gemini 3.5 Flash, Omni, Spark, and Antigravity 2.0

Google announced Gemini 3.5 Flash at I/O 2026 with a 1M-token context window, 65k max output, four thinking levels, and pricing of $1.50 per 1M input tokens and $9.00 per 1M output tokens.

Why it matters: HKR-H/K/R all pass: this is a Google I/O model-and-product bundle with concrete context, output, thinking-tier, and pricing facts. It has same-day relevance for Claude, OpenAI, and coding-agent competition, so it clears P1.

AI HOT (Curated Pool)May 20, 08:00

Unsustainable Subsidies

Google, OpenAI, and Anthropic diverged on model pricing: Gemini 3.1 Pro is priced at $2 input and $12 output, GPT-5.5 at $5 and $30 after a short subsidy, and Claude Opus 4.7 stayed at $5 and $25.

Why it matters: HKR-H/K/R all pass, but this is Tom Tunguz commentary on pricing rather than a primary model release. The concrete price spread makes it featured, not must-write.

AI HOT (Curated Pool)May 20, 06:49

Gemini Omni Supports Video Creation With Personal Likeness and Voice

Gemini Omni lets users create digital-avatar videos using their personal likeness and voice, and the avatar can generate videos without uploading an image each time; the post does not disclose pricing, regions, or launch timing.

Why it matters: HKR-H/K/R pass: personal avatar video is clicky, reusable identity is a concrete mechanism, and voice/likeness raises creator and safety stakes. Price, regions, and launch timing are not disclosed, keeping it near the featured floor.

AI HOT (Curated Pool)May 20, 06:40

Gemini 3.5 Flash price rises sharply as Google plans broad rollout

Google released Gemini 3.5 Flash at I/O with $1.50 per million input tokens and $9 per million output tokens, making it 3x and 6x the previous model’s pricing, while adding roughly 1 million input tokens and about 65,000 maximum output tokens.

Why it matters: HKR-H/K/R all pass: a Google model update with a sharp pricing twist and concrete token costs. It stays below p1 because the body only gives price and rollout intent, not capability deltas, benchmarks, or context window.

AI HOT (Curated Pool)May 20, 06:08

Gemini launches personal AI agent and Daily Brief

Gemini added Gemini Spark and Daily Brief: Spark acts as an always-on personal AI agent across Gmail, Google Docs, and Slides after user authorization, while Daily Brief is available to Google AI subscribers in the U.S. aged 18 or older.

Why it matters: HKR-H/K/R all pass: Google is adding Gemini Spark’s authorized actions across Gmail, Docs, and Slides, plus Daily Brief eligibility for US 18+ AI subscribers. This is a same-day Google agent product update.

TechCrunch · AIMay 20, 05:25

You Can Now Talk to Your Gmail Inbox, as Seen at Google I/O 2026

Google expanded Gmail’s AI Inbox with conversational voice search, letting users ask Gemini to find details buried in email. The RSS snippet does not disclose rollout scope, supported languages, pricing, latency, or the retrieval mechanism behind Gmail search.

Why it matters: HKR-H/K pass: a Google-scale Gmail voice inbox feature is concrete and clickable. HKR-R is weak because rollout, language support, pricing, and retrieval mechanics are not disclosed.

AI HOT (Curated Pool)May 20, 05:12

Google launches new AI search box with multimodal interactions

Google launched an AI search box based on Gemini 3.5, combining AI Overviews and AI Mode into one AI search experience that supports multimodal multi-turn queries across text, images, files, and video, with global availability on desktop and mobile.

Why it matters: HKR-H/K/R all pass: a Google Search entry-point update with Gemini 3.5, multimodal file/video queries, and AI Overviews/AI Mode integration. The source is thin, so it lands at the lower end of must-write.

The Verge · AIMay 20, 05:00

Google’s AI Future Demands Trust — and Your Personal Data

Google presented Gemini Spark, Daily Brief, and expanded Gmail AI inbox access at I/O 2026; the Verge snippet says these tools depend on large amounts of personal information, but the post does not disclose detailed data-handling terms.

Why it matters: HKR-H/K/R all pass, but the body gives product names and a personal-data dependency without data-handling details. Google I/O makes it featured, not a same-day must-write.

Take

Google’s risky move is not an always-on Gemini Spark; it is making Gmail, Calendar, and tasks the model’s front door. I/O 2026 names Gemini Spark, Daily Brief, and Gmail AI inbox, spanning event planning, daily summaries, custom to-dos, and personalized replies. Every feature needs private context. The available RSS snippet gives no retention rules, training exclusions, enterprise-domain boundaries, or human-review conditions.

I don’t buy the “useful products earn trust” story here. Google’s moat is distribution through Gmail and Calendar, which OpenAI and Anthropic cannot easily copy. Its exposure is the same surface area. Microsoft has at least kept pointing Copilot buyers to M365 tenant boundaries; Google’s disclosed pitch looks more like occupying the default workflow first and asking users to supply the trust later.

Financial Times · TechnologyMay 20, 04:47

Google to Release Smart Glasses and Add AI Agents to Search Engine

Google will release smart glasses and add AI agents to its search engine; CEO Sundar Pichai says features powered by a new Gemini model will narrow the gap with Anthropic and OpenAI, while the RSS snippet does not disclose specs, launch timing, or pricing.

Why it matters: HKR-H/K/R all pass: Google is moving Gemini agents into Search and smart glasses, a core entry-point product story. Missing specs, pricing, and timing keep it below the top band, but it fits the 85–94 must-write range.

Take

Google is betting on owned surfaces, not a clean Gemini win over Claude or OpenAI. The disclosed moves are specific: agents inside Search, plus smart glasses. The snippet gives only Pichai’s claim about closing the gap; it gives no specs, timing, pricing, context window, or task boundary for the agents.

I don’t buy the “catch-up” framing yet. Google’s durable advantage over the last year has been default distribution: Search, Android, Chrome, Workspace, YouTube. OpenAI and Anthropic won developer and prosumer mindshare through ChatGPT and Claude; Google can push agents into workflows users did not actively choose. The glasses angle smells like an Android XR distribution test. Ray-Ban Meta already showed that camera, voice, and lightweight notifications land faster than a general assistant story.

AI HOT (Curated Pool)May 20, 04:25

Smarter Google AI Edge Gallery: MCP Integration, Notifications, and Session Continuity

Google AI Edge Gallery adds experimental MCP support on Android, letting Gemma 4 coordinate external data sources including Google Workspace and Google Maps; the update also adds scheduled notifications and persistent chat history for faster restoration of long-session context.

Why it matters: HKR-H/K/R all pass: Google’s developer update adds experimental MCP, notifications, and session continuity to AI Edge Gallery. It is a mid-weight product update, not a model release or major capability launch.

AI HOT (Curated Pool)May 20, 03:25

Google Tensor ML SDK Beta Released

Google released the Tensor ML SDK beta, letting developers convert, compile, and run PyTorch or TFLite models on Pixel 10 TPUs through LiteRT, with a model library containing more than 100 classic and generative AI models, including Gemma 3.

Why it matters: HKR-K is strong: the post gives a concrete Pixel 10 TPU workflow and a 100+ model library. HKR-H/R clear the featured bar, but this is a beta developer SDK rather than a flagship model or major consumer launch.

Take

Google is making a distribution play for on-device AI, not proving that phones now run serious LLM workloads. Tensor ML SDK Beta ties PyTorch/TFLite conversion, compilation, deployment, inference, Play Feature Delivery, AI Packs, and CPU/GPU fallback into LiteRT. That plumbing matters because edge ML usually dies on packaging, runtime support, and device fragmentation, not on a single demo latency number.

The 100+ model garden sounds broad, but the hard examples are Gemma 3 1B, Function Gemma 270M, and EmbeddingGemma 300M. That is useful for local actions, semantic features, camera tricks, and speech flows. It is not a cloud-agent replacement. Apple keeps its on-device path tighter and more closed; Qualcomm’s NPU story still leaves developers stitching vendor pieces together. Google’s advantage is the LiteRT + Hugging Face + Play delivery loop. Performance, power draw, and Pixel 10 install base are not disclosed, so the victory lap is premature.

AI HOT (Curated Pool)May 20, 03:19

Gemini Omni launches with physical reasoning and multimodal generation

Google launched Gemini Omni video generation for global AI Plus, Pro, and Ultra subscribers, integrating it with Gemini app, Google Flow, and YouTube Shorts, while the RSS snippet says the model combines intuitive physical reasoning with Gemini’s historical, scientific, and cultural knowledge.

Why it matters: HKR-H/K/R all pass: this is an official Google Gemini video-generation launch with named tiers and product surfaces. Details are thin—no benchmarks, pricing, or physical-reasoning tests—so it stays at the low end of the 85–94 band.

TechCrunch · AIMay 20, 03:15

Google takes a page from Meta, announces audio-powered smart glasses at I/O 2026

Google announced “audio glasses” at I/O 2026, letting users issue voice commands across its apps and services, including Gemini; the RSS snippet does not disclose price, launch timing, or hardware specifications.

Why it matters: HKR-H/K/R pass: Google announced Gemini-linked audio glasses at I/O 2026, a credible AI-hardware platform move. Missing price, launch date, and specs keep it in the low featured band.

Take

Google is claiming the glasses entry point before showing a real product. The RSS snippet only says “audio glasses” support voice commands and can call Gemini plus Google apps. It gives no price, launch date, chip, camera setup, battery life, weight, or distribution plan. For AI-device teams, those missing fields matter more than the Gemini name; glasses fail first on comfort and battery, not model branding.

Meta Ray-Ban already proved the lower-friction path: no display, voice-first, camera plus earbuds behavior. Google is walking that same lane, but with better native tools on paper: Android, Maps, Gmail, Calendar, and Assistant/Gemini hooks. The wild part is that Google should have owned this category years ago. Without hardware specs or shipping timing, this is still an I/O ecosystem marker, not a Meta problem yet.

AI HOT (Curated Pool)May 20, 03:07

Google releases Gemini Omni multimodal generation model

Google released Gemini Omni, a multimodal generation model that combines image, video, and text inputs to generate videos grounded in Gemini’s real-world knowledge; the post does not disclose model size, pricing, or availability.

Why it matters: HKR-H/K/R all pass: Google’s Gemini Omni adds combined image, video, and text inputs for video generation. Missing parameters, pricing, and rollout timing keep it at the low end of the must-write band.