Tanmay Garg

@ChatGPTapp tata, bye bye - FillApp, FillMate, Fillify, Fill Hero, AgentFillAI, VeloFill, WebFill, FillAI.

Paperwork is better when you can just talk through it. With Images in ChatGPT and voice mode, you can upload a form, say what to fill in, and get back a completed version.

@thsottiaux @aisuperapp log

@aisuperapp Linear or log scale?

AI Super App Rankings 1. Codex 2. Claude Desktop App 3. Cursor 4. Gemini

@thsottiaux Bruh,

Dark magic. Codex.

Education is the foundation of all progress. It currently have all world knowledge to guide you but currently it's limited by post trained text output upto a particular length. Can it output a small book as per user requirement with proper index, chapters, diagrams, animations, like a interactive book, curated to the level of the user? Would need a different post train and different harness

what problem do you most hope AI will solve in the future? maybe we can help!

7/7 So yes, Omni is cool. But the real Google I/O story is not video. It’s permission. Who gets to act for you, where they act, and how much of your digital life they can touch without feeling creepy. That’s the agent race.

6/7 My actual question: who gets trusted first? The model that gives the best answer? Or the ecosystem that already has your inbox, files, calendar, browser, and payment context? That answer probably decides more than benchmark charts.

5/7 This is why Gemini Spark matters more than it looks. A 24/7 personal agent sounds like marketing until you connect it to: calendar email browser search payments shopping workspace files then it becomes less like a chatbot and more like an operating layer.

4/7 OpenAI wants ChatGPT to become the super app. Anthropic is winning a lot of serious workplace trust. Google’s counter is different: put the agent inside Search, Chrome, Android, Gmail, Docs, shopping, coding, YouTube. boring distribution beats beautiful demos.

3/7 The model race is still real. But “best model” is becoming a temporary advantage. The more durable fight is: who owns the place where AI actually does the work?

2/7 Look at the pattern: Search agents Daily Brief Gemini Spark Universal Cart Antigravity That’s not “we made a smarter chatbot.” That’s “AI now gets a workspace, memory, tools, and permission to act.”

1/7 🧵 Google I/O takeaway: everyone is yelling about Omni video. fair. it’s flashy. but I think the bigger move is quieter: Google is trying to turn the whole internet into an agent runtime.

Small thing I think people are underweighting: The public video leaderboards are already crowded: Seedance, Kling, Veo, Sora-style models all trade blows on first-render quality. Omni’s interesting bet is different. Not “can it make a good 10 sec clip?” More: can it survive the 5th edit?

@GoogleDeepMind So the benchmark I’d love to see is not just blind A/B: “which clip looks better?” It’s: give every model the same starting ref make 5 natural-language edits then judge whether the final video still holds together That’s probably where the next real moat is.

That’s the part creators actually care about. Same character after a few changes. Same scene logic. Camera still makes sense. Physics doesn’t quietly break. The reference you gave it still matters. A lot of AI video looks great on attempt #1, then falls apart when you try to direct it.

We’re dropping Gemini Omni: our first step towards a model that can create anything from anything - starting with video. It combines Gemini’s intelligence with our generative media systems - representing a leap forward in world understanding, multimodality, and editing 🧵

@bcherny @DavidKPiano Opus dies after 2 request in normal plan

@DavidKPiano Hey, Boris from the team here. What can we do better?

Been using Codex much more than Claude Code lately It's wild that Claude became popular *because* it had the best-in-class models for coding... and all it had to do was hold the lead

@ChatGPTapp For PPTs also pls

ChatGPT is now available as an add-on in Excel and Google Sheets. It can help analyze messy data, write formulas, update spreadsheets, and explain what it’s doing along the way—without leaving your spreadsheet. Powered by GPT-5.5. chatgpt.com/apps/spreadshe…

@sama Put the same in codex, fine tuned for coding, it doesn't understand tech stack specific terms. Like how do you even pronounce shadcn

people are really starting to use voice to interact with AI, especially when they have a lot of context to dump. GPT-Realtime-2 comes to the API today; it is a pretty big step forward. (we are working on improvements to voice in chat.)

And they said we have hit a wall

@Dimillian If it connects to unity and creates the same, that would be game changer

I've asked Codex for simple polygons but it seems like it cooked way beyond that goal