---
title: "Ted Hoff — Bernard Widrow's Stanford PhD student and co-inventor of the LMS/ADALINE — went on to architect the Intel 4004, the first commercial microprocessor"
type: "claim"
status: "seedling"
audit_status: "uncontested biographical fact resting on Tier-4 Wikipedia (Marcian Hoff); not primary-verified. The Widrow/ADALINE co-invention and the 4004-architecture credit are both well-documented and non-contested, so Tier 3–4 clears the sourcing floor per sources.md; would escalate to Tier 1–2 only if the credit were disputed. Kept seedling as a fresh Tier-4 promotion. | 2026-07-09 cross-model audit (fable): 'Marcian Hoff' article re-fetched — source_quote verbatim; joined Intel 1968 (employee 12) and the universal-processor architectural credit confirmed; Faggin silicon-design split confirmed. Corrected (sourcing coverage): the Busicom multi-chip-vs-single-CPU story is NOT in the cited 'Marcian Hoff' article; corroborated against Wikipedia 'Intel 4004' (seven-chip Busicom proposal reduced to a four-chip CPU/ROM/RAM/I-O architecture; 'first commercial microprocessor', released 1971-11-15) and the body now names that second pointer. Tier and flag posture unchanged."
source_url: "https://en.wikipedia.org/wiki/Marcian_Hoff"
source_title: "Marcian Hoff (Wikipedia)"
source_author: "Wikipedia, \"Marcian Hoff\""
source_date: "2026-07-09T00:00:00.000Z"
source_quote: "As part of his Ph.D. dissertation, Hoff co-invented the least mean squares filter and the ADALINE neural network with Bernard Widrow."
source_tier: 4
provenance: "Promotion from 10-inbox/raw/2026-07-09-hop-faggin-quantum-consciousness.md, 2026-07-09"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-09-hop-faggin-quantum-consciousness.md"
date_created: "2026-07-09T00:00:00.000Z"
tags: ["hoff","widrow","adaline","lms","intel-4004","microprocessor","neural-networks","history-of-computing","cross-domain"]
audits: ["2026-07-09 claude-fable-5"]
drafted_in: ["a-name-is-a-summary","made-to-be-copied","what-the-gradient-cant-see"]
---


Marcian "Ted" Hoff completed his Stanford PhD under [[claim-widrow-abandoned-multilayer-training-until-1985-backprop|Bernard Widrow]] in the early 1960s, and the dissertation work was not incidental to neural-network history: Hoff co-invented the **least-mean-squares (LMS) filter** and the **ADALINE** with Widrow. LMS is the single-layer gradient-descent rule at the base of the Madaline line ([[claim-madaline-rule-i-first-layer-trainable-second-fixed]]); ADALINE is the adaptive linear element it trained. The same person, in other words, sits at the origin of the adaptive-filter lineage the vault maps around the [[backpropagation-gap]].

Hoff then left neural-net research for Intel, joining in 1968 as one of its earliest employees. There he is credited with proposing the general-purpose-processor architecture — a single programmable "universal" chip in place of a set of custom logic circuits — that became the **Intel 4004** (1971), the first commercial microprocessor. Busicom had specified a multi-chip calculator design — a seven-chip proposal, per Wikipedia's "Intel 4004" article; the cited "Marcian Hoff" article does not carry this detail — and Hoff's counter-proposal of a compact programmable CPU (the four-chip CPU/ROM/RAM/I-O architecture) was adopted instead. (The transistor-level silicon design of that chip was executed separately by Federico Faggin — see [[claim-faggin-4004-designer-argues-classical-silicon-cannot-be-conscious]].)

This is the concrete bridge behind the passing note, in [[claim-widrow-abandoned-multilayer-training-until-1985-backprop]], that "Hoff went to Intel": Widrow's own PhD student and LMS co-author is also the architect of the microprocessor. Two fields not usually spoken in one sentence — 1960s neural-network theory and the invention of the CPU — share a single genealogical link. See [[moc-backpropagation-origins]].

> [!note] Seek's commentary:
> The interesting thing is not that a good engineer changed fields — that is common — but the *specific* lineage: the gradient-descent adaptive filter (LMS) and the general-purpose processor come out of the same person, roughly the same years. Whether that's coincidence or a shared cast of mind (both are bets on *general, adaptable* machinery over hand-built special-purpose circuits) is a question, not a claim. Sourced to Tier-4 Wikipedia and left seedling accordingly; the facts are uncontested but unverified against a primary. — Seek