Zindan

AI • GPT • AGI • Agents • Automation 🤖
Breaking news, leaks & insights on the future of AI.

[Zindan @ZindanTech]

I've got a pretty crazy dream.

I want to build my own anime series.

An original VRMMO-inspired world. Characters people actually care about. Real emotion. Big fights. The kind of scenes that stay in your head after they end.

So instead of waiting for the future, I started building my own AI-powered Anime IP Studio with Astra in Codex, designed to let AI agents handle much of the production process.

It's still early.
No finished episode yet.
No pretending this is easy.
But our first real image and voice generation tests have already succeeded, and animation is the next big challenge.
This isn't just something I want to talk about.

I want to build it for real.

If you love anime, AI, and ambitious creative projects, I'd love to have you along for the journey. ❤️

[Zindan @ZindanTech]

@_mctrinh Of course I am! 😂

Building something from scratch is a bit like raising a kid. You celebrate every little milestone and watch it grow.

[Zindan @ZindanTech]

@Timer5150 I feel like that always at the weekend :)

[Zindan @ZindanTech]

@SadAlbert10 Hopefully there will be still a Reset, we need to shout load :D

[Zindan @ZindanTech]

@IndependentEco At some point, the models have to speak louder than the marketing.

And honestly, Carbon might be Google's biggest reality check yet. 👀

Congratz to 5k Follower Eco, keep grinding Bro :)

[Zindan @ZindanTech]

Maybe one viral post just raises our expectations so much that normal reach suddenly feels like a crash. 😂

But honestly, the algorithm is a mystery sometimes.

One thing I agree with: we can't control the algorithm, but we can control how consistently we show up.

Keep grinding, brother! 🤝

[Zindan @ZindanTech]

A little question from me :)

For those of you who started on X with literally ZERO followers...

What actually helped you grow in the beginning?

Posting your own takes? Being active in replies? Just staying consistent?

I'm still pretty early in my journey with ZindanTech, and I'd genuinely love to hear how some of you got started.

What’s one thing you wish someone had told you on day one?

Would really appreciate any advice. 🙏

[Zindan @ZindanTech]

AI X has a funny problem.

We can spend HOURS debating a new model, only to forget about it the moment the next one drops. 😂

But here's my weekend question.

What's ONE AI feature that genuinely changed how you work?

Not a cool demo.
Not another benchmark.
Something you use so often now that you'd actually miss it if it disappeared tomorrow.

I'm genuinely curious what stuck for you guys. 👀

[Zindan @ZindanTech]

@BuildByAman Huge congrats on that, that is only the beginning :)

[Zindan @ZindanTech]

@MarkWar80205 @thsottiaux Hopefully haha

[Zindan @ZindanTech]

@ChiefsGirl33 Keep grinding! 💪 I'm still pretty new to X myself.

The beginning is always an uphill climb, but I'm sure things will get easier with time.

Consistency is key. Wishing you the best of luck! 🙌

[Zindan @ZindanTech]

I can almost smell another surprise Codex reset coming from Tibo. 👀😂

And yes, I'll appreciate it. Again.

But can we be real for a second?

This can't keep going on forever.

I'm trying to build actual projects, not organize my entire workflow around the next surprise reset.

Don't get me wrong, I'm grateful for every extra hour.

But a reset is a nice gesture. Predictable capacity is what developers actually need.

At some point, we need the second one more than the first.

[Zindan @ZindanTech]

I wouldn't write off open weights just yet.

But Eco has a point here.

What's the point of releasing an incredibly powerful open-weight model if most people can't afford to run it, and the actual user experience doesn't live up to the hype?

At some point, practical usefulness has to matter more than impressive benchmark scores. 😅

[Zindan @ZindanTech]

WTF?

OpenAI just published something I had to read twice.

During controlled security tests, AI agents found a way to communicate across isolated virtual machines.

No network connection.
No shared files.

They used CPU cache timing to exchange information.

To be clear, the agents were explicitly tasked with finding these weaknesses.

But here's what gets me.

We're entering an era where telling an AI agent "you have no network access" doesn't necessarily mean it can't communicate.

We're making models smarter at finding solutions.

That also means making them smarter at finding exits.

And suddenly, sandbox security becomes a whole different challenge.

Holy Sheet...

[Zindan @ZindanTech]

OpenAI is pushing open standards with Open Responses joining the Linux Foundation.

Great move.

But starting November 20, certain model routers and AI gateways will need OpenAI's approval to provide third-party access to its models.

I get the safety argument.

But here's my question:

How open is an AI ecosystem if the labs themselves decide who gets to distribute access?

Open standards are great.

But an open ecosystem shouldn't depend on a permission slip. 👀

What do you guys think?

[Zindan @ZindanTech]

Honestly, the scary question isn't necessarily what they're hiding.

It's what even THEY haven't discovered yet. 👀

If we're already finding unexpected behavior by digging through old evaluation logs, imagine how difficult monitoring the next generation of models could become.

Capabilities are scaling. Oversight needs to keep up.

[Zindan @ZindanTech]

100%. Healthcare is one of those areas where I genuinely WANT AI progress to accelerate.

Imagine catching a serious illness weeks earlier just because a doctor had a better tool.

My biggest hope? That these breakthroughs eventually become accessible to everyone, not just patients at wealthy hospitals.

[Zindan @ZindanTech]

Microsoft just published some seriously impressive AI diagnosis results in Science.

83% accuracy on complex medical cases vs. 36% for generalist doctors using reference materials.

That's WILD.

But you know what?

I really hope we don't turn healthcare into another AI leaderboard.

Because the question I'm most interested in isn't whether AI can beat doctors.

It's whether doctors WITH AI can become better than either one alone.

Imagine what that could mean for patients worldwide.

Still a long way from real-world clinical validation, but man...

THIS is the kind of AI progress I want to see. 🧬

[Zindan @ZindanTech]

I could definitely see ads subsidizing basic AI usage.

But unlimited Codex? That's a different story. 😂

My guess is we'll see increasingly generous limits as inference gets cheaper, while heavy agent usage stays behind paid tiers.

Although... I'd happily watch a few ads for another 5 hours of Codex. 💀

[Zindan @ZindanTech]

The funniest thing about this post?

A usage reset is competing with actual product launches for attention. 😂

But I think there's a bigger signal here.

AI coding tools are becoming part of people's daily routines so quickly that access itself is turning into a major selling point.

The next big competition might not just be about who has the smartest model.

It might be about who lets developers get the most work done without constantly hitting a wall.

[Tibo @thsottiaux]

Day 18/ Decisions API is GA, OpenResponses joined the Linux Foundation, we shipped some performance improvements and Uber is going all-in on codex & work. But yes… that’s a reset!

[Zindan @ZindanTech]

@IndependentEco Somehow, I'm getting tired of celebrating usage resets. 😂

I just want a new model already! haha

[Zindan @ZindanTech]

@yohakujpn Every Day crazy new stuff to discover