Kumazori 💙

my tabs don't rule me. i rule the tabs.
AI tools, productivity experiments and random rabbit holes. probably rewatching The Nice Guys instead

[Kumazori 💙 @Kunazomi]

@github this is the kind of sdk update that makes building with ai ay less intimidating🙃having the agent loop and mcp setup handled for you means you can spend more time building the actual app ..

[Kumazori 💙 @Kunazomi]

11 seconds for a full lip sync video is seriously fast!!😅 being able to use one photo and an audio clip makes this way easier to play around with too

[fal @fal]

H3 Max Lip Sync is now live on fal.

Upload one photo and one audio clip and get back a lip-synced video in seconds, with natural expression and movement, in any language.

In our evals, H3 Max Lip Sync ranks #1 for both quality and speed, with a median generation time of just 11 seconds.

Try it now on fal.

[Kumazori 💙 @Kunazomi]

@Arm 300k+ creators and around 90% being first timers really stands out! lowering the production barrier could mean a lot more people getting their stories made��

[Kumazori 💙 @Kunazomi]

i found google’s new cc and this is basically ai for the stuff nobody wants to organise 😅

it can pull together things from gmail, calendar and your drive into one briefing..with shared context across tasks

honestly, this is the ai i’d use for keeping life admin from becoming a second job..

[Kumazori 💙 @Kunazomi]

@DiscussingFilm world war z 2 is actually happening!!! brad pitt is back and edward berger is directing… definitely watching this

[Kumazori 💙 @Kunazomi]

i fell down a forecasting rabbit hole today 😂

metaculus lets people predict future events using probabilities instead of simple yes/no answers

so instead of saying this will happen… just think there’s a 70% chance..

that little shift makes you think about uncertainty way differently!

[Kumazori 💙 @Kunazomi]

this is a very cool direction for ai 🤖 using it to speed up scientific research while letting people focus on the questions and decisions that reaallyyy matter..

[Google Research @GoogleResearch]

"The best way to predict the future is to invent it."

In a deep dive with VP, Google & GM, Google Research @ymatias on The Google Research Podcast, @sineadbovell explored how AI is transforming science & interfaces:

🔹 Ambient AI & GenUI: Interfaces that adapt in real time to how you think and learn.
🔹 Virtual Labs: Tools like AI Co-Scientist turning multi-year hypothesis generation into 3-day sprints.
🔹 Scientist as Architect: AI gives every student & researcher the resources they need to focus on asking better questions and exercising critical judgment.

AI isn't replacing the researcher — it's an amplifier for human curiosity and judgment. 🔬⚡

👉 Tune in to our pilot episode to hear the full conversation: https://t.co/WZ8xdR0Q9b

[Kumazori 💙 @Kunazomi]

@FlowbyGoogle tools really can change how you work... the right one can take an idea that MIGHT work to an idea that WILL work!!⚡️

[Kumazori 💙 @Kunazomi]

@CultureCrave a new cat species after 100 years?? and the fact that someone found it as a kitten thinking it was just a weird looking house cat makes this even better😂

[Kumazori 💙 @Kunazomi]

@sterkinekor this is bringing back so much nostalgia 😂 i’m definitely gonna need to see this one 🪄

[Kumazori 💙 @Kunazomi]

this is where ai research gets weird fast 😳if models start improving the systems that improve models....

the research loop could get A LOT tighter.. the hard part is knowing how much of that compounds in practice

[Dwarkesh Patel @dwarkesh_sp]

Once AIs automate AI research, should we expect some crazy RSI that is analogous to the explosion we’re seeing in math right now for well scoped (but still crazy ambitious) problems?

@polynoamial and I debate this question out:

[Kumazori 💙 @Kunazomi]

@NewsFromGoogle @GoogleLabs okay.. this is kinda useful
getting the family stuff out of everyone’s heads and into one shred agent could save a ridiculous amount of back and forth😅

[Kumazori 💙 @Kunazomi]

i found microsoft's new ai playbook...this part got me 🫠

they looked at 100+ internal ai projects and basically found that throwing a better model at everything isn't gonna magically fix it!

sometimes the actual problem is the messy stuff around it ..like bad data or missing contxt ..maybe the ai is just not having enough info to do the job properly

so before adding another ai tool to the pile, i'd look at what’s actually slowing the workflow down first...

give the ai better context nd clean up the process THEN worry about which model you're using

ey, that might save you from having 17 ai tabs open for no reason 😂

[Kumazori 💙 @Kunazomi]

pika just made ai video creation a lot less annoying 😉

instead of having to figure out which model...tool or prompt you need for each part, its new workflow can handle different parts of the video for you

you can start with an idea, then let it work through the image.. movement.. sound and edits!!

honestly, that could save a LOT of messing around when you’re trying to turn one idea into an actual video....there's less jumping btween tools nd more time actually making he thing

[Kumazori 💙 @Kunazomi]

@XBusiness @1x_tech @Kenny___Rose ngl seeing the neo robot in an actual campaign is kinda cool😄 feels way more fun than just another product announcement..

[Kumazori 💙 @Kunazomi]

this makes so much sense... catching a bad camera move or weird object placement before the final render could save so much time and money😮‍💨

[Rohan Paul @rohanpaul_ai]

The expensive mistake in AI video usually happens before rendering: the scene itself was wrong.

A better model doesn't fix a bad camera path or broken spatial layout.

GPT-6 Astra is now live on OnSolo, and the Whitebox Video Expert makes the most sense to me.

I can describe a scene, Astra reasons through the space and motion, and the system gives me a whitebox pass before the final render.

So I can validate blocking, object placement, and camera movement while everything is still cheap to change.

[Kumazori 💙 @Kunazomi]

60 minutes to deploy an ai data center with no concrete modifications is a crazzy idea!!

..if the setup actually works at scale, that could make getting compute online way simpler

[TBPN @tbpn]

Rune builds modular AI data centers that produce no noise, no pollution, and can be deployed in 60 minutes without any concrete or modifications.

Co-founder & CEO William Layden breaks it down:

[Kumazori 💙 @Kunazomi]

@NVIDIARobotics 68.1 tokens on a jetson is actually nuts⚡️double the throughput on the same hardware is a nice upgrade for local ai !

[Kumazori 💙 @Kunazomi]

@netflix 12 hours is actually a tease😕 netflix really knows how to make you wait for the good stuff!!!

[Kumazori 💙 @Kunazomi]

i went down a very weird ai road...and somehow ended up thinking about power grids😂

emerald ai just launched an alliance around making ai data centres more flexible with their electricity use!

so...when the grid gets stressed some ai workloads could slow down so the important stuff keeps running

the more ai gets deployed, the more "how smart is the model?" is gonna sit next to "how much electicity does this thing need?"