A few questions: • Why do I need a phone call when I’m already on my laptop? Isn’t that extra friction? • Would you actually book the flight/hotel Dot picked without manually checking everything yourself price, payment, cancellation policy, baggage, etc.? • If I still have to review screenshots, verify details and make the final decision, how much time is actually being saved? • Would love to see how this handles real failure cases too. • Also, how was Wi-Fi the main culprit when it was able to answer your 2nd query in the live demo?

Never ever ordering from @zomato @zomatocare again. These mf are not at all helpful when there’s an issue with an order. Shut down your company if you don’t have a single human available to interact with customers and all you can provide are AI generated responses.

A day in the life, in three minutes. Made with @claudeai Sonnet 5.5 in one shot, still needs improvement

First priority of viswaguruland, fight for local languages and send your children abroad and english speaking schools/colleges.

Cute. Not sure it justifies taking up valuable usage limits though.

We went from "AI will write all our code" to "AI will clean up the code AI wrote" The circle is complete.

"Beyond excited to work closely with these groups to figure out how AGI can serve them" lol

We’re reaching the point where “what’s the best AI model?” is becoming a useless question. The better question is: best at what? Coding, agents, reasoning, speed, cost, long-context, multimodal… the rankings are changing constantly.

i think dot didn't work well here #devday

Two things went wrong during the @OpenAI #devday demos: • During the dot demo, the dot wasn’t responding. • In another demo where Codex was being used, voice chat wasn’t working at all. And now I can’t even find these clips on the official stream anymore. 👀 But I recorded them live while the stream was happening.

A few questions: • Why do I need a phone call when I’m already on my laptop? Isn’t that extra friction? • Would you actually book the flight/hotel Dot picked without manually checking everything yourself price, payment, cancellation policy, baggage, etc.? • If I still have to review screenshots, verify details and make the final decision, how much time is actually being saved? • Would love to see how this handles real failure cases too. • Also, how was Wi-Fi the main culprit when it was able to answer your 2nd query in the live demo?

Reverse-engineering a game binary is impressive, but why assume that means the same model can compromise banking systems, Aadhaar infrastructure, or airline/flight systems? A game binary is distributed to the user and runs locally. Critical banking, Aadhaar and airline infrastructure is generally server-side, with authentication, authorization, network segmentation, encryption, monitoring, rate limits and other controls. And "Claude refuses to tick Im not a robot so we’re protectet" is also a weak security argument. Model refusals add friction, but they aren't the security boundary. An unrestricted OSS model could potentially make browser automation, vulnerability research and attack workflows cheaper and more scalable. The real question is, can AI reliably automate the full chain from vulnerability discovery -> exploit development -> attack-chain construction ->execution? If yes, that’s a serious cybersecurity problem. But why jump from "AI can reverse-engineer a game" to "banking, Aadhaar or flight systems are compromised" without demonstrating those intermediate steps?

From my personal experience it's really good to catch these things and call them out thanks to being powered by Astra but we'll see how we can better demo some of those situations

What are the top 3 use cases where Dot has actually saved you significant time or effort? And how is it meaningfully different from Meta @Muse or @bot curious to know

Oh, sure, let me give ChatGPT all my financial info. But seriously, how is this sensitive data stored, used, shared, and deleted? Is it ever used for model training?

Thanks for the reply! What I meant by failure cases was more about real-world failures during the actual workflow for example, a flight becoming unavailable, the price changing before checkout, a payment failing, a cancellation policy not matching what Dot understood, or Dot making an incorrect assumption about one of the constraints. I’d be really interested to see how Dot detects these situations, recovers from them, and keeps the user informed without requiring them to manually redo the whole process. That’s the kind of failure handling I was referring to.

Every quote and reply promoting this is starting to feel like an ad campaign without anyone actually saying it’s sponsored.@elonmusk