Alexander Whedon

Introducing SubQ - a major breakthrough in LLM intelligence. It is t...

Name: Introducing SubQ - a major breakthrough in LLM intelligence. It is t...
Uploaded: 2026-05-05T14:00:15.000Z
Duration: 1 min 24 s
Channel: Alexander Whedon
Description: Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is:

Lançamento

Plataforma: X
Duração: 1:24
Idioma falado (inferido): inglês
Visualizações: 13,5 mi
Corte mediano: 1.14s
Rosto em cena: 60%
Quadros escuros: 78%

Full video footage is a Pro feature. The stills are free — the source film is $10/mo.

Upgrade to Pro

Dark-Mode Explainer Interview

medium effort

Steal this
Pillarbox the interview so the black side bars are the exact same black as your full-frame graphic cards. The talking head and the animation then live in one continuous dark canvas, and you can cut between them every second without the frame ever flashing.

A single seated presenter interview is chopped into ~1-second slices and interleaved with full-frame, near-black motion graphics that visualise each claim. The presenter footage is pillarboxed (vertical crop centred in a 16:9 frame) so the black side bars match the black graphic cards, letting the edit cut between talking head and animation without any visual jolt. Burned-in two-or-three-word captions run along the bottom of both layers, tying the whole thing together.

Opening move

Cold open on a full-frame near-black graphic — a titled benchmark bar chart, thin serif/light sans headline top-left, bars rising into frame with one bar accent-green against neutral cream — before any face appears. The claim is shown as a chart first, so the viewer is reading a result rather than being introduced to a person, and the black background instantly signals 'product launch' rather than 'vlog'.

Pacing

Relentless: 17 detected shot changes across 85s, median 1.14s between cuts. Assume roughly one cut per sentence fragment, with several sub-second flashes inside the animated grid sequence (three of the sampled frames sit within one second of each other around the 38s mark, i.e. a hard cut or a fast animated push on nearly every beat). Alternate blocks — two or three quick presenter sizes, then a graphic slab, then back. 78% of frames read dark, so let the graphics dominate screen time (18 no-face shots vs 27 presenter shots) and keep the face on screen only about 60% of the time.

Graphics and type

Two-tone system: near-black background, cream/off-white as the neutral fill, one accent colour per idea (green for the hero bar, yellow highlight strokes on documents, a single saturated blue disc, one red square). Headlines in a light serif or thin sans, top-left, generously letter-spaced, with a small subhead beneath. Chart labels tiny and unbold — deliberately understated, even slightly overlapping. Captions are burned in, centred at the bottom, medium-weight white sans, no box or stroke, only two or three words per card so they change on almost every cut. Presenter footage is graded low-key and slightly desaturated: cool green background, warm skin, crushed shadows, white shirt kept just under clipping. Vignette or natural falloff darkens the frame edges so the pillarbox blends into the graphics.

Cut at the same speed

Lit and framed like this

More launch