Q2B26 Copenhagen | Saeed Jahromi, Director of Quantum Research, Multiverse Computing

Value of the Information & Strength of the Argument

The talk delivers substantial value by connecting two cutting-edge fields—quantum computing and AI—through a common mathematical framework. The argumentation is structured and logical: it first establishes the theoretical basis of tensor networks, then demonstrates their utility in quantum simulation and validation, and finally applies the same principles to compress AI models. Real-world examples (IBM’s kicked Ising, Fermi-Hubbard) strengthen the case, and the inclusion of a hybrid quantum-classical LLM pipeline shows a concrete path forward. However, the argumentation is one-sided, focusing on successes and not addressing limitations or potential failures of tensor network methods in certain regimes. The speaker also does not provide comparative analysis with other compression techniques, leaving the viewer with a promotional impression rather than a balanced scientific assessment.

Scientific Rigor, Source Quality, Title Accuracy

The speaker references several external experiments and internal developments, but no detailed citations or publications are given in the talk. The only explicit external source is the conference website (q2b.qcware.com), which provides context but not primary research. The talk lacks a critical evaluation of the cited work, and claims such as ‘3,000 times faster’ are presented without references. The title accurately matches the content, though it does not hint at the specific focus on tensor networks. No comments were available for analysis.

The title accurately reflects the content: a specialist talk on tensor networks bridging quantum computing and AI, delivered by the Director of Quantum Research at Multiverse Computing.

7/10

The speaker is a research director at a quantum computing company, presenting both conceptual frameworks and specific technical results. Claims are supported by references to known experiments (IBM kicked Ising, Fermi-Hubbard) and internal benchmarks, but they are not independently verified in this talk. The promotional context reduces objectivity, yet the technical content is coherent and consistent with established tensor network literature.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:30Introduction to tensor networks as a bridge between quantum and AI.
• 02:25Definition of tensors and tensor networks, and their origin in quantum many-body physics.
• 05:00Use of tensor networks for quantum computing simulation and validation of quantum advantage claims.
• 07:35Examples: IBM kicked Ising experiment and Fermi-Hubbard simulation, showing classical tensor network methods can match or surpass quantum results.
• 10:10Transition to AI: high training costs of large language models, and the need for compression.
• 11:30Introduction of Compactified, a tensor-network-based AI compressor, with a 100B model compressed 90% with minimal accuracy loss.
• 13:00Deployment of compressed AI on edge devices like Raspberry Pi, demonstrated at a conference.
• 14:10Hybrid quantum-AI approaches: injecting quantum circuits or quantum neural networks into LLM architectures enables running them on quantum computers.
• 16:00Example of an end-to-end hybrid quantum-classical LLM pipeline run on IBM quantum hardware in Spain.
• 17:30Wrap-up: tensor networks as a unifying tool, with future AI support for quantum error mitigation.

• Q2B Copenhagen Conference — The talk was presented at Q2B26 Copenhagen; the website provides general conference information.

• Tensor network simulation of IBM's kicked Ising experiment — Peer-reviewed work independently showing that tensor network methods can efficiently simulate the experiment, supporting the speaker's claim.

• Critique of quantum advantage claims in annealing systems — A study that challenges some quantum advantage claims for D-Wave systems, aligning with the speaker's caution but also potentially generalizing the concern to all such claims.