Sirui Lu

Physics graduate student @TU_Muenchen @MPI_quantum

[Max Welling @wellingmax]

Our book on Generative AI and Stochastic Thermodynamics can be pre-ordered with a 20% discount until July 31 2027. (All proceeds from the authors will be donated to the African Institute for Mathematical Sciences).

[Sirui Lu @ThreeDeerQ]

@kevinweil @thefriley There might be a small but subtle difference - many everyday chores I talk with ChatGPT I don’t mind for training as long as they are anonymous. But for research contents academics would actually want proper attributions while it is being anonymized?

[Sirui Lu @ThreeDeerQ]

@BigBreakfastLob @texra_ai you mean the original pictures ? Yes the would ok too. There should be some initial human input like the latex template with chapters division and so on, but once the norm is set the agent can learn/generalize

[Sirui Lu @ThreeDeerQ]

@boazbaraktcs For this kind of application think that you might like the agents inside @texra_ai :)

[TeXRA.ai @texra_ai]

Use our https://t.co/kxcAig0K93 and do this for any directions everyday! research is now 10x more fun and exciting

[Sirui Lu @ThreeDeerQ]

@Zhuoc3Chen curious to understand the conceptual differences between your “new scaling laws” with what obtained in https://t.co/rGqEsmnuds

[Sirui Lu @ThreeDeerQ]

@kevinweil @OpenAI Maybe you’d be interested in our co-discovery of many many new quantum error correction codes outside of the normal classes with multi-agents powered by GPT-5 too :) https://t.co/ii89fQa3Mq

[Antoine Tilloy @AntoineTilloy]

At the @DeutschesMuseum, Frank Verstraete explains how Ignacio is really the Renaissance man of the second quantum revolution.

(Viva Quantum conference for the 60 years for Ignacio Cirac in Munich)

[Sirui Lu @ThreeDeerQ]

@JiaqingJ Very nice result & nicely illustrated boundary!

[Sirui Lu @ThreeDeerQ]

@kevinweil if any of you from OpenAI actually read what was screenshotted, you would know in this case o3 mini did not help more than (1) write out a given hamiltonian and its transfer matrix which is by definition (2) a block diagnolization of a small matrix...

[Sirui Lu @ThreeDeerQ]

Let the world champion final matches decided by who plays better on the board, not by whether players wear jeans, not by where to put the captured stones (as long as it is reasonable), not by the referees who ruthlessly interrupting at the mostcritical moment. #LGCupWorldGoKings

[Sirui Lu @ThreeDeerQ]

which is not cited by newer references that connects RLHF etc to inference/KL control

[Danilo J. Rezende @DaniloJRezende]

That's a very old idea eg see this SL review (https://t.co/aK3r5KXLx3) for older refs (Todorov, Kappen etc)

[Sirui Lu @ThreeDeerQ]

@keenanisalive What would be the meanings of the spatial locations of particles? On both the left and right the densities are fixed at a specific point.

[Nicolas Delfosse @nic_delfosse]

Ever wonder what is the probability of a fault in your classical processor? On Astra supercomputer they estimate about 6 faults per node per day where each node contains 56 processors 2GHz. That's about one fault every 1e^14 cycles.

[Sirui Lu @ThreeDeerQ]

@YT59529321 no one disagree QC is very powerful; But it is a different computer good and bad in many ways. Shouldn't we be humble and embrace the progresses in sister fields and strive to identify the areas where QC have the most advantage, and accept if it turns out to be limited?

[Sirui Lu @ThreeDeerQ]

@YT59529321 The point is that DFT gives extremely good results for many materials we care about. Use AI methods to find better functionals etc just make it better. The scope of where QC ab-initio calculations (which is a long way to go) are useful are thus scarce.

[Sirui Lu @ThreeDeerQ]

@YT59529321 @dallairedemers What did you find in the article in your view would be unbalanced (or unfavor of QC?) All of the cited research are incredibly solid, and presents good results already today.

[Sirui Lu @ThreeDeerQ]

@YT59529321 whether it is new or not is not relevant to the discussion. One can also say QC is not new. The important point is that ML inspired methods for simulating q. sys/materials are advancing rapidly

[Sirui Lu @ThreeDeerQ]

@YT59529321 Variational methods can help measure quantum entropies of 50 sites. Would QC ever be able to do that without running the experiments for 10 years?