nick

telepathy @conduitpbc | meta-level watcher | 这个世界是个草台班子

[nick @n6cholas]

@GeorgeLutas1 @NotTuxedoSam https://t.co/u6nwc3Z204

[nick @n6cholas]

super proud of what we’ve been up to the last few months. if you’re curious about BCIs and ML, check it out

[Lydia (in SF) @LydNot]

so @conduitpbc spent 6 months running a basement lab 20 hours/day to collect the largest neuro-language dataset ever built—~10,000 hours from thousands of people.

the wild part isn’t the model. it’s the operations.

turns out:

•⁠ ⁠noise matters < 4k hours; after that, scale eats it

•⁠ ⁠dynamic overbooking beats every recruitment strategy

•⁠ ⁠a 4-pound helmet is fine if you cut custom polygonal padding

•⁠ ⁠rewriting the backend cut marginal cost/hour by ~40%

•⁠ ⁠people will happily wear multimodal headsets if the LLM talks to them, not at them

•⁠ ⁠zero-shot decoding works on people the model has never seen:

thought → text:

“the room seemed colder” → “there was a gentle breeze”
“do you have a favorite app?” → “any favorite robot?”

the future of thought-to-text isn’t sci-fi.

it’s logistics, padding foam, airflow pipes, and a booking algorithm.

full post: https://t.co/Y8harECGU8

[Lydia (in SF) @LydNot]

so @conduitpbc spent 6 months running a basement lab 20 hours/day to collect the largest neuro-language dataset ever built—~10,000 hours from thousands of people.

the wild part isn’t the model. it’s the operations.

turns out:

•⁠ ⁠noise matters < 4k hours; after that, scale eats it

•⁠ ⁠dynamic overbooking beats every recruitment strategy

•⁠ ⁠a 4-pound helmet is fine if you cut custom polygonal padding

•⁠ ⁠rewriting the backend cut marginal cost/hour by ~40%

•⁠ ⁠people will happily wear multimodal headsets if the LLM talks to them, not at them

•⁠ ⁠zero-shot decoding works on people the model has never seen:

thought → text:

“the room seemed colder” → “there was a gentle breeze”
“do you have a favorite app?” → “any favorite robot?”

the future of thought-to-text isn’t sci-fi.

it’s logistics, padding foam, airflow pipes, and a booking algorithm.

full post: https://t.co/Y8harECGU8

[nick @n6cholas]

@blingdivinity fair - do you have raw cot from asking another model eg “what do you think ‘marinade’ means to claude”?

[nick @n6cholas]

@merowing_ @iamtrebuh @browsercompany except that arc did have ai integration and everyone was happy… then they killed it anyway

[nick @n6cholas]

@NVIDIAGeForce GeForce Day

[nick @n6cholas]

@interaction poke didn’t take its meds today

[nick @n6cholas]

@pancakeufo @SmileForSalmon @tomhaerter @Victor_Patru @cursor_ai but crucially we cannot buy .edu domains for non-university use in the united states, which is not true of .it — in fact, more broadly, educational institutions as a whole in the united states are barred from .edu unless they qualify as higher education (i.e. college/uni)

[nick @n6cholas]

@ToivoMattila @robinebers @YashwantSi22011 @dev_11217 @cursor_ai it was a grok model that went unreleased most likely

[nick @n6cholas]

@kimmonismus no, it doesn't. this chart assumes max token length for **every prompt** and a figure for energy consumption that was speculated before gpt5 was even released.

conservative estimate: ~0.4wh/prompt, 2.5b prompts/day --&gt; ~0.37twh annually

2.2% of what you posted

[nick @n6cholas]

@theo unfortunately very real phenomenon