Neuralink PRIME, Webgrid (Neuralink, 2024)
Measured by Neuralink (PRIME Study, participant Noland Arbaugh) · Neuralink PRIME Study progress update (2024)
Inputs
The measured or assumed values behind the calculations, each with its source.
- t = 30
- Webgrid grid dimension: the standard board is 30×30 = 900 cells (Neuralink's Webgrid source: BPS = log2(t²−1)·f/60, f = net correct clicks/min).
- B = 8.01 bits/s
- Participant Noland Arbaugh's Webgrid record (BPS), reported by Neuralink. First-ever session: 4.6 BPS. The often-quoted ~10 BPS is the able-bodied-mouse reference on the same board, not a participant result.
- f = 49 selections/min
- Net correct selections/min, back-derived from 8.01 BPS on the 30×30 board: f = 8.01·60 / log2(899) ≈ 49 (one target ~every 1.2 s).
Strictest ITR
Each scoring method is an upper bound on the channel, so the headline is the strictest (smallest) one for this entry. Use the score selector on the home page to view any single method across entries.
-
Webgrid geometry → movement difficulty
30×30 board: cell width W = S/30; random targets → mean amplitude A ≈ 0.52·S, so A/W ≈ 15.6. ID = log2(A/W + 1) = log2(16.6) ≈ 4.06 bits/movement.
S = board span; 0.52 is the mean distance between two random points on a square. The target is cued, so each correct selection is a pointing movement carrying the Fitts index of difficulty (≈4.06 bits), not the log2(899) ≈ 9.81 bits the Webgrid metric credits as a 1-of-900 choice.
-
Re-credit the reported achieved bitrate onto the Fitts basis
ITR = B × ID / log2(N − 1) = 8.01 × 4.06 / 9.81 = 3.32 bits/s
Neuralink publishes only the achieved bitrate B = 8.01 bits/s and no independent selection timing, so the Fitts figure is that same measured throughput re-credited per movement at the Fitts ID instead of the full grid entropy. This is a unit re-crediting of the reported score, not an independent measurement. Validation: the able-bodied ~10 BPS on this board re-credits to ~4.1 bits/s, matching MacKenzie's measured mouse throughput (4.5 bits/s), so the cursor lands just below an able-bodied mouse.
What counts as a bit depends on the action space. The number of distinguishable actions and how likely each one is are design choices of the task, not the sensing hardware. The same modality can present a fixed set of targets, a set pruned per step by a grammar or language model, or a continuous control space. Each of these changes how many actions are live and how the probability mass is spread, and therefore the information per selection. Read the action space below before comparing headline numbers across entries.
Action space
What the user can produce at each step, and how those options are distributed.
- Structure
- Continuous control space
- Size
- Continuous
- Prior
- Uniform: all actions assumed equally likely
- Notes
- A 2D cursor controlled by an N1 intracortical array, measured on Webgrid (a cued target-acquisition task). The headline BPS is a Webgrid task metric, not an ISO-style Fitts throughput: it assigns log2(t²−1) ≈ 9.81 bits for each correct selection on the 30×30 board, while the atlas pointing convention uses the movement difficulty, log2(A/W+1) ≈ 4.1 bits, because the target is cued before the movement. The reference number here is therefore a Fitts-style throughput on Neuralink's own grid; the Webgrid BPS is kept as a secondary, as-reported figure. Since Webgrid's BPS = log2(t²−1)·f at the net-correct rate, it is shown as the achieved-bitrate grid score (scoreType nuyujukian) for comparison with the other secondary task metrics.
Comparability The strictest bound here is Fitts throughput: the index of difficulty, log₂(A/W + 1), per movement. Directly comparable to the other continuous-pointing entries (mouse, trackball, stylus, gaze and the cursor BCIs). Set against the text entries (keyboards, spellers, speech) it crosses methods: both report bits/s, but one measures movement difficulty and the other text information, so compare within the family first.
Other score types
Bounds the atlas keeps out of the default strictest headline: as-reported figures, alternate task conditions, or raw-channel ceilings that shouldn't win the headline by default. Each still carries a score type, so the home-page selector ranks this entry on it when you choose that type. Read its derivation before comparing across entries.
-
Webgrid achieved-bitrate metric
30×30 board → N = 900 cells. The Webgrid score credits log2(N − 1) = log2(899) ≈ 9.81 bits per net-correct cued selection (Neuralink's own Webgrid formula, a log2(N−1) variant of the achieved-bitrate metric; Nuyujukian et al. 2015, which introduced the metric, uses log2(N)).
-
Authors' reported score (taken as reported, not re-derived)
B = 8.01 bits/s. Noland Arbaugh's Webgrid record, reported directly by Neuralink (first session 4.6 BPS; the often-quoted ~10 BPS is the able-bodied-mouse reference on this board, not a participant result).
Neuralink publishes only the BPS score, with no net-correct selection rate or per-trial timing. There is no more-primary quantity to derive this from: it is an author-reported figure. Dividing out the per-selection credit implies ~8.01/9.81 ≈ 0.82 selections/s (~49/min), but that rate is a consequence of the score, not an independent measurement, so this is not a forward calculation.
Source
- Authors
- Neuralink (PRIME Study, participant Noland Arbaugh)
- Publication
- Neuralink PRIME Study progress update, 2024