At Sequoia’s AI Ascent 2026 (YouTube published May 28, 2026), Neuralink president and co-founder DJ Seo (Dongjin Seo) framed the company’s long-horizon aim as a higher-bandwidth interface between brains and AI — including early “neural foundational model” fine-tunes on neural data at roughly 20 participants — while still treating current work as restoration of lost function. That is executive/interview packaging. It is not a peer-reviewed model card, not proof of an exocortex product, and not a redo of the Sep 6 careers-page hiring-language brief. Age of the video is disclosed up front.

Sequoia Capital — Neuralink’s DJ Seo: Inside the Race to Connect Brains and AI — Sequoia’s AI Ascent 2026 interview with Neuralink president and co-founder DJ Seo, hosted by Shaun Maguire (published May 28, 2026 — an older video, disclosed up front). These are interview claims; patient counts are anchored to Neuralink’s own Telepathy update.

At a glance

Interview ≠ model card. Seo’s Sequoia session is the right packaging for Neuralink’s brains-and-AI ambition. It is not a published neural foundational model paper, and it is not proof an exocortex product exists.

  • Confirmed (interview): ~20 / over-20 patient scale framing; neural FM / transformer fine-tune language; restoration-first benefit-risk; bandwidth / raw-intent long horizon.
  • Confirmed (company blog): 21 participants announced in “Two Years of Telepathy”; Telepathy and VOICE described in investigational clinical context.
  • Fence: Conference ambition ≠ FDA filing; hiring-page NFM brief already covered Sep 6 — this lead is Seo’s interview packaging.

What Seo actually put on the record

Speaking with Sequoia partner Shaun Maguire at AI Ascent 2026, Seo treated AI as central to Neuralink’s origin story: the bottleneck between human input/output and machine capability. Maguire put “over 20 human patients” in front of the audience; Seo said that even at “20 or so participants,” the company is seeing early results from a “neural foundational model” — state-of-the-art LLMs or transformer networks fine-tuned on neural data, with some patterns he called counterintuitive.

He also widened the AI meeting point beyond “think into a chatbot.” In some horizon, he said, AI becomes an “exocortex,” and the real breakthrough would be computing on raw intent rather than translating thought into keyboard, mouse, and language. Useful executive thesis. Still a thesis.

Restoration first — augmentation later, if ever

Seo repeatedly framed current work as restoration of lost function for people whose potential benefit can justify surgical risk. On elective enhancement for otherwise healthy people, he returned to benefit-risk, payment, and approval — not a shipping consumer product. House voice keeps that distinction visible even when the conversation drifts to Matrix-style concept transfer.

Anchor patient counts to company writing

Neuralink’s January 28, 2026 update Two Years of Telepathy announces 21 participants (“Neuralnauts”) and walks through Telepathy use cases plus the VOICE trial’s conversational-speed ambition. Use that blog to ground “~20 patients” interview vibes in company-stated enrollment language — without converting either source into a peer-reviewed outcomes table.

Why this is not the Sep 6 careers brief

AI Shift News already covered Neural Foundation Models language from Neuralink’s careers page. Today’s flagship differentiates on primary packaging: a Sequoia-hosted interview with the president/co-founder, including scale, caregivers, Blindsight write-vs-read framing, and the restoration-to-augmentation benefit-risk discussion. Blindsight talk here is not a human vision-restore receipt (Seo’s end-of-year hope was preclinical framing in-session). VOICE speech restoration is covered in today’s other Neuralink Sunday flagship.

Bottom Line

Seo’s Sequoia tape is the clearest public executive packaging of Neuralink’s brains-and-AI ambition at ~20-patient scale. Report the claims. Disclose the May 28, 2026 upload date. Keep the proof bar on peer-reviewed model cards, FDA filings, and shipping products — none of which this interview is. That is the fence: brains and AI as interview thesis — not a published neural FM receipt.

Sources