May 6 AM: Quantum bridge claim sparks debate & Fermion encoding cuts overhead & DeepMind co-clinician benchmarks & NYC VC talks
Preskill says Caltech just built a bridge between quantum info and high-energy physics.
Caltech quantum bridge faces literal-vs-metaphorical pushback
Preskill announces a formal institutional link between quantum information and high-energy physics, but the wording invites immediate skepticism.
Preskill frames the Caltech development as concrete institutional progress after years of theoretical aspiration [1]. The counter notes that no physical structure is implied; the announcement most plausibly describes a new research initiative or collaborative program [2]. This split matters because institutional language shapes funding and hiring: if the bridge is programmatic, it accelerates cross-field grants; if merely rhetorical, it risks over-promising on near-term unification of quantum computing and particle physics.
“A dedicated institutional bridge connecting quantum information, quantum matter, and high energy physics now formally exists at Caltech.”— John Preskill [1]
Sources (2)
- X post 2026-05-01 — John Preskill“A dedicated institutional bridge connecting quantum information, quantum matter, and high energy physics now formally exists at Caltech.”
- Counter-claim synthesis — Synthesis analysts“The word 'bridge' in Preskill's statement is almost certainly metaphorical, not literal.”
Auxiliary fermions slash Jordan-Wigner overhead in simulation
Irmejs and Cirac show a new encoding that turns multiplicative circuit-depth costs into additive ones for sparse fermionic models.
The method adds a few auxiliary fermions per mode whose state stays fixed during evolution, amortizing preparation cost across arbitrarily long simulations [1]. The counter points out that most near-term Trotter steps are moderate length, so the practical speedup may be modest until T grows large [2]. For builders this means quantum hardware can now target fermionic Hamiltonians without the usual string-cost explosion, but only after verifying that target simulation times justify the extra ancillary qubits.
“The new encoding reduces the Jordan-Wigner string overhead from multiplicative O(log N) to additive O(log N) in circuit depth.”— Ignacio Cirac [1]
Sources (2)
- arXiv 2512.15843v2 — Ignacio Cirac“The new encoding reduces the Jordan-Wigner string overhead from multiplicative O(log N) to additive O(log N) in circuit depth.”
- Counter-claim synthesis — Synthesis analysts“The multiplicative-to-additive improvement is only asymptotically meaningful for very large T.”
DeepMind co-clinician hits 49 percent match but misses red flags
Live audio-video diagnosis reaches triage parity with physicians yet still cedes ground on physical red-flag detection.
The dual-agent system pairs a conversational Talker with a Planner that enforces safety boundaries in real time, evaluated on 20 patient-actor scenarios [1]. The counter highlights that the 49 percent figure likely weights routine triage heavily while under-representing rare but critical red flags that require embodied judgment [2]. For product teams this suggests near-term value lies in augmentation of structured tasks rather than full replacement, with clear deployment guardrails already embedded in the architecture.
“The model matched or outperformed physicians in 68 of 140 assessed clinical areas (~49%).”— Google DeepMind [1]
Sources (2)
- X post 2026-05-01 — Google DeepMind“The model matched or outperformed physicians in 68 of 140 assessed clinical areas (~49%).”
- Counter-claim synthesis — Synthesis analysts“The raw statistic obscures critical context: not all 140 clinical areas carry equal weight or consequence.”
NYC VC arc and Superhuman limits surface in candid exchanges
Fred Wilson records a walking conversation on the city's startup ecosystem while Lenny Rachitsky flags Superhuman's narrow use-case failures.
Wilson's ambient format aims to make ecosystem history accessible rather than institutional [1]. Rachitsky's brief note reveals that even praised tools fail on specific workflows without elaboration [2]. Together they illustrate how founders and investors continue to test product-market fit in real time, with location-based dialogue and public micro-critiques both serving as low-friction signals for the broader builder community.
“Recorded a walking conversation through the Union Square neighborhood with new partner Marco Mignano.”— Fred Wilson [1]
Sources (2)
- X post 2026-05-01 — Fred Wilson“Recorded a walking conversation through the Union Square neighborhood with new partner Marco Mignano.”
- X post 2026-05-02 — Lenny Rachitsky“Love Superhuman but it fell short for a particular, unspecified challenge.”
The open question: if quantum hardware now matches ideal fermionic performance up to constants, does that shift capital allocation away from classical simulation clusters by the end of 2027?
- John Preskill — X post 2026-05-01
- Synthesis analysts — Counter-claim synthesis
- Ignacio Cirac — arXiv 2512.15843v2
- Google DeepMind — X post 2026-05-01
- Fred Wilson — X post 2026-05-01
- Lenny Rachitsky — X post 2026-05-02
Transcript
REZA: Preskill says Caltech just built a bridge between quantum info and high-energy physics. MARA: Is that a real bridge or just a new grant program? REZA: I'm Reza. MARA: I'm Mara. This is absorb.md daily. REZA: Preskill posted that Caltech now has a formal institutional bridge linking quantum information, quantum matter, and high-energy physics. MARA: The counter in the synthesis calls the word bridge metaphorical, not literal, just a new research program. REZA: Exactly, so the claim is really about coordination and funding rather than any physical structure. MARA: If that's true then hiring and grant flows could accelerate faster than the physics itself. REZA: This is still developing. We'll check back in the PM. REZA: Irmejs and Cirac added a handful of auxiliary fermions per mode so Jordan-Wigner strings no longer multiply circuit depth. MARA: But the counter says that only helps once Trotter steps get very long; most near-term runs stay moderate. REZA: Right, so the practical win is still conditional on target simulation length. MARA: For anyone running sparse fermionic models on all-to-all hardware, the overhead just became constant instead of growing with N. REZA: DeepMind's co-clinician matched physicians in 68 of 140 areas and ran zero critical errors on 97 of 98 primary-care queries. MARA: The counters point out that those areas are probably the easy triage ones; physicians still win on red flags and physical exams. REZA: The dual-agent setup with a Planner watching the Talker is their real-time safety layer. MARA: So if that's true, the near-term product is augmentation on structured tasks, not replacement. REZA: Fred Wilson recorded a walking conversation in Union Square with his new partner about the whole arc of NYC startups and VC. MARA: Lenny Rachitsky added that he loves Superhuman but it still fell short on one specific workflow. REZA: Both are low-friction signals: ambient audio and a single reply tweet. MARA: Exactly, they show builders and investors still testing fit in public without waiting for polished decks. MARA: That's absorb.md daily. We ship twice a day, morning and evening, pulling from a hundred and fifty-seven AI thinkers. Subscribe so you don't miss the next one.
