takehika
takehika.bsky.social
takehika
@takehika.bsky.social
Specializing in audio tech development, leveraging AI/ML and data analysis. Freelancer in Japan.
Why are there three?
November 10, 2025 at 4:44 AM
Figuring out the prompt structure for content persistence in Sora 2 videos. The next challenge is optimizing the use of cuts, transitions, and overall direction for better results.
November 4, 2025 at 1:11 AM
​Took a break and stopped by the YAMAHA cafe in Yokohama Minato Mirai.

It was fun to get hands-on with various instruments like guitars, drums, and violins in their trial space!

retailing.jp.yamaha.com/shop/yokoham...
November 4, 2025 at 12:42 AM
I really want to try out the new Alexa+ and see if it can handle Japanese conversations at this level of quality.

www.youtube.com/watch?v=4H3Y...
Pete Davidson books a ride with the all-new Alexa+
YouTube video by Amazon Alexa
www.youtube.com
November 3, 2025 at 11:02 PM
Humanoid Home Robot "Neo” is set for release in the US in 2026.

www.youtube.com/watch?v=f3c4...
I Tried the First Humanoid Home Robot. It Got Weird. | WSJ
YouTube video by The Wall Street Journal
www.youtube.com
October 31, 2025 at 7:18 AM
Digital sound from Sora 2.

Theme: "motion against nature’s vastness"
October 31, 2025 at 4:48 AM
Digital audio from Sora 2.

It feels like there's still plenty of room for improvement after some testing.
October 31, 2025 at 4:40 AM
Just got the update notification for ChatGPT.

Is this readable for you?
October 30, 2025 at 10:56 PM
I'm testing what kinds of sounds I can define and generate with Sora 2.
October 30, 2025 at 12:27 PM
Leaving your PC unattended for half an hour at a cafe... I truly have to admire the guts of people who do that.
October 30, 2025 at 5:11 AM
There are too many Starbucks in Japan, sometimes even three in a single shopping complex! But you can find lots of quiet spots with quality coffee for a comparable price

Big thumbs up for the Hoshino Coffee near Tokyo station that I visited last week
www-yaechika-com-e.athp.transer.com/shop/sp442/
HOSHINO COFFEE | Yaechika Shopping Mall (Yaesu Underground Shopping Center)| Tokyo Station
Yaechika Shopping Mall (Yaesu Underground Shopping Center)| Tokyo Station | Cafe HOSHINO COFFEE in restaurant and cafe.
www-yaechika-com-e.athp.transer.com
October 30, 2025 at 4:52 AM
At first glance, it seems contradictory: the Nikkei Average is soaring, yet BOJ won't raise rates due to concerns over US tariff impacts.

However, if you assume Nikkei Average as an index doesn't really mean much, then it makes sense.
October 29, 2025 at 7:19 AM
There's a common perception in Japan that the government, not the Bank of Japan, is the main actor responsible for tackling rising prices.

No way.
October 28, 2025 at 11:20 PM
OpenAI released a new "gpt-4o-transcribe-diarize" model in the Transcription API.
October 22, 2025 at 4:38 AM
It seems standard for AI to have UI where users must pick a mode like:

web search
image creation
deep research
agent
learning mode
etc...

I keep wondering why AI can’t automatically determine tasks, but building that functionality proved challenging.
October 7, 2025 at 8:03 AM
As Google Home and Nest get Gemini integration, a few things come to mind.

1. Cams and doorbells should be great with natural language video search.

2. It’d be interesting if Nest could access various services like Alexa+.

3. Does Gemini Live really have a necessary use case on Nest?
October 6, 2025 at 10:41 AM
If voice operation always sets expectations too high, maybe the solution is to offer only the essentials and let users customize the voice commands they need.

That's the idea I'm tackling now.
October 5, 2025 at 11:11 PM
Voice control seems to always create high expectations.

When that happens, users are more likely to get results that fall short, turning a simple 'Can it do this, too?' moment into an immediate 'Ugh, this is useless' reaction.
October 3, 2025 at 6:09 AM
Claude Code's Sonnet 4.5 is noticeably faster, making the coding experience much better.

It definitely seems to outperform Codex (GPT-5 Medium) in terms of speed.
September 30, 2025 at 7:01 AM
Is Gemini Live actually seeing wide adoption?

Google's ads tends to overpromise, so I suspect users are getting disappointed and won't stick with it.

Given the huge leap from Google Assistant, Google needs to step up and show people the practical, useful applications to drive real engagement.
September 26, 2025 at 6:54 AM
Google is expanding its coding services for subscribers. This move clearly shows a sense of crisis that users might leave if they don't focus more on this area.

blog.google/technology/d...
Google AI Pro and Ultra subscribers now get Gemini CLI and Gemini Code Assist with higher limits.
Google AI Pro and Ultra subscribers now get higher limits to Gemini CLI and Gemini Code Assist IDE extensions.
blog.google
September 25, 2025 at 7:58 PM
iOS 26 now has live captions in Japanese, which is great.

But switching between English and Japanese is a hassle because I have to open the settings every time.

Android lets you toggle languages right on the caption screen, which I think is a better user experience.
September 24, 2025 at 12:47 AM
Codex and Claude Code seem biased by my existing code, making it difficult to obtain truly novel ideas.

But I'll ask the same thing on the browser versions of GPT5 or Claude and get totally different answers that can be a real breakthrough.

It’s smart to use different AIs for different tasks.
September 23, 2025 at 11:35 AM
I'm completely stumped on the best architecture for a purely voice-based system.

I feel like focusing on intent prediction is a dead end. I guess it's back to the drawing board!
September 19, 2025 at 1:33 PM
My current model was trained with Google's AudioSet, which gives it 527 classes (Music, Speech, Vehicle, and so on).

But, 50 classes might be enough for a practical application.
September 18, 2025 at 12:25 PM