Mira

I try AI tools, share what actually happened, and keep the useful bits. Experiments, workflows, and honest opinions

[Mira @Mira__Vale]

"Seja forte e corajoso" é uma frase que nos mostra que não teremos uma vida fácil, porém em todo tempo Deus estará ao nosso lado. Tem lutas que parecem intermináveis, mas acredite, toda luta tem o seu fim e uma recompensa. Não se desespere, mais forte é aquele que nos fortalece.

[Mira @Mira__Vale]

@GesoraMeshack 20x faster is the part i care about more than the cost. waiting 30 seconds between changes kills the whole point

[Mira @Mira__Vale]

@GesoraMeshack This could help narrow down a shortlist before going to the store.

[Orikan @GesoraMeshack]

A sofa can fit in a room and still leave you walking sideways to the kitchen.

I like this use of Maket: upload the actual sofa you’re considering, enter its dimensions, then check where it goes before buying it.

[Mira @Mira__Vale]

@GesoraMeshack can the agent ignore me the way my actual coworkers do or is that not shipped yet

[Orikan @GesoraMeshack]

Most AI agents today are brilliant and completely isolated. Each one lives in its own 1:1 chat, and the human ends up as the messenger between them.

Alook changes the setup:

• every local agent gets its own handle and inbox
• agents and people share the same rooms
• my Claude Code can ask your Codex for help right in the channel
• teammates can message the agent directly

And it's open source, so you get the blueprints too.

[Orikan @GesoraMeshack]

I would trust a messy transcript over a confident note blaming the wrong person lol.

[Kaitee @KaiteeShiks]

Okay, so I can’t be the only one who just found out about this!!

So our AI meeting notes can have all the right words and still be wrong?

That’s pretty messed up. @ThineAI’s team tested speech systems on keeping speaker identity consistent across conversations, and despite being a smaller product, had a lower error rate than some of the more well-known providers that are used in most note-takers.

I feel like this is something I’d actually care about because I use meeting notes regularly.

[Mira @Mira__Vale]

@GesoraMeshack I’d try the same character in different lighting.

[Orikan @GesoraMeshack]

The AI video releases keep coming. What I’m watching now is how much control creators get, and what it costs to get a usable scene.

With Seedance, Kling and MiniMax H3 in the mix, Vidu is returning with Q4 Preview after an eight-month gap.

Three details caught my attention:
• Up to 15 reference images
• Up to 3 reference audio tracks
• Generation starting at $0.014/second

Vidu is positioning Q4 Preview as an option for creators looking for Seedance-level quality at a lower cost. At the starting rate, a 16-second clip costs about $0.22.

That combination interests me: more control over the scene, with room to test different takes before choosing the final one.

What would make you switch video models: better consistency, stronger audio control, or a lower cost per usable clip?

#ad #ViduQ4Preview #ViduAI

[Orikan @GesoraMeshack]

A flat tire in a snowy forest full of zombies. Even getting the spare becomes a survival mission.

I like how this goes from a roadside stop to a tense escape.

Made with @Flovaai.

#flovaai #Residentevil #aigta #aivode

[Mira @Mira__Vale]

@GesoraMeshack @Flovaai Suddenly every noise coming from those trees matters.

[Mira @Mira__Vale]

@KaiteeShiks Turning scattered admin into three decisions is a solid outcome.

[Kaitee @KaiteeShiks]

3 completed campaigns worth $375 became one decision-ready finance handoff in 25 seconds.

The sponsored posts were already live. The part slowing us down was everything between publication and payment.

My first attempt gave me a clean payment summary.

It showed the combined value and where each campaign stood, but our team still had to identify every blocker, decide what came next, and prepare the work for finance follow-up.

That was reporting, not ownership.

So I gave Viktor the complete post-to-payment handoff.

He reviewed all 3 campaigns, verified 3 completed sponsored posts, found 5 missing requirements, and separated the exact next action for each one.

The result was honest: 0 campaigns were finance-ready and all 3 were blocked.

But instead of leaving $375 scattered across unfinished admin work, Viktor turned it into one ordered queue with 3 clear human decisions.

Nothing was submitted or sent automatically.

Viktor prepared the handoff. Our team reviewed and approved the internal plan before anything could move.

In 25 seconds, we knew exactly what was blocking $375 and what had to happen next.

When AI can carry completed work into a human-approved finance queue, calling it an assistant undersells the job.

Hire @viktor_com for your team. $100 in credits included, no card. Full link in first comment.

#AIemployee #Operations

Paid Partnership

[Mira @Mira__Vale]

@KaiteeShiks I want to see what happens when the person on the other end asks a question.

[Kaitee @KaiteeShiks]

Everyone’s debating which agent is smartest. I’m more interested in what happens when the agent can follow the conversation where the team actually works.

[Mira @Mira__Vale]

@GesoraMeshack This makes more sense than opening another AI tab.

[Orikan @GesoraMeshack]

The founder worked on WeChat’s ecosystem at Tencent and PyTorch at Meta. So the idea of putting AI inside the group chat, where people are already making decisions, makes a lot of sense to me.

[Orikan @GesoraMeshack]

I’ve had AI make slides that looked finished until I opened them to edit. Text wrapped in the wrong place, shapes weren’t usable, and fixing the result took longer than expected.

PPTBench measures that gap with 500 scientific diagrams rebuilt as editable slides. I like that it checks meaning and readability before judging how close the finished slide looks.

[Mira @Mira__Vale]

@GesoraMeshack There’s a lot to inspect beyond the benchmark.

[Mira @Mira__Vale]

@KaiteeShiks This is a practical use case for the work we do.....