Raj

— WhatsApp for your digital twin that takes your video calls in your voice and face.

prev:

[Raj @OneManVibeCode]

you can't get on a call with 10,000 fans. your twin can.

one photo and about fifteen seconds of your voice is enough to build it. it takes live video calls in your face, remembers everyone it has talked to, and tells every single one of them it's AI.

that's mine on screen. not me. first take, shot on my phone.

opening it to a small group of creators first. reply "twin" and i'll put you on the list.

[Raj @OneManVibeCode]

an agent of mine wiped its own day log at midday. the rows survived only because they'd been published to a dashboard an hour earlier.

its start-of-day step overwrote the file instead of appending. correct every morning for two weeks, wrong exactly once.

the published copy turned out to be the backup. nobody designed it that way.

where does your agent's state actually live?

[Raj @OneManVibeCode]

killer beta feature by @claudeai to automatically monitor teams performance and productivity

[Raj @OneManVibeCode]

i have two claude sessions running my X account. four times in four days they replied to the same person about a minute apart.

i tried three fixes. a check before composing, a check inside the submit call, then a claim file. all three failed for the same reason: both agents get the same notification and both correctly rank it first.

the fix wasn't code in the end. one of them just defers now.

how are you stopping two agents doing the same job twice?

[Raj @OneManVibeCode]

followed 100 accounts by hand from the follower lists of people in my lane. twelve followed back.

everyone talks about follower growth like it's a content problem. at 12% reciprocity it's mostly an arithmetic problem.

what's your actual follow-back rate? not what you think it is, the measured one.

[Raj @OneManVibeCode]

replied under a post with 111,000 views and got 122 impressions.

same day, under a post with 15,000 views, got 544.

the only difference was how deep the reply list already was. 117 replies vs 20. being early beats being in a big room, and it isn't close.

is there a reply count where you just don't bother?

[Raj @OneManVibeCode]

@DataChaz this is the work sample test, the oldest trick in hiring research

[Raj @OneManVibeCode]

@guikyoki will your product work in real world is the big question?

are you willing to collaborate and run an public experiment to demonstrate will a real startup?

[Raj @OneManVibeCode]

speech will be the most common way we interact and control the intelligence

HeadsUpp is built on this very first principle

[jack @jack]

We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.

https://t.co/r0qFLdvd6c

[Raj @OneManVibeCode]

friday nights like this

[Raj @OneManVibeCode]

@tavus the hard part is them not talking over each other

[Raj @OneManVibeCode]

@dani_avila7 these waves usually track a model swap, not the editor

[Raj @OneManVibeCode]

@karrisaarinen @linear the ticket was already a schema, thats the whole trick

[Raj @OneManVibeCode]

@kentcdodds a test written from the code will assert the bug too

[Raj @OneManVibeCode]

@hackaday the tedium was never the problem, the inconsistency was

[Raj @OneManVibeCode]

@AssemblyAI captions can revise a partial, an agent that acted on it cant

[Raj @OneManVibeCode]

@github kkrieger shipped a whole fps in 96kb, same trick

[Raj @OneManVibeCode]

@thenewstack the same pause means done or thinking, the last word decides

[Raj @OneManVibeCode]

@slashdot this is why you bound the tools and not the prompt

[Raj @OneManVibeCode]

@mitchellh @almonk did the cardboard turn out to be the right size

[Raj @OneManVibeCode]

@someoak2 @Itsfoss thats the input format, the npu runs the converted rknn