When the Trillion Parameter Consortium holds its 2025 all‑arms convention and exhibition, TPC25, in San Jose on July 28–31, computational scientists, system architects, coverage leaders, and AI stakeholders from analysis, business, and authorities will sort out one query: how can trillion‑parameter frontier fashions be constructed, shared, and ruled for on a regular basis science?
Over three days, TPC25 attendees will dive into tutorials on constructing brokers and a arms‑on hackathon. Plenary audio system reminiscent of Argonne’s Rick Stevens and RIKEN’s Satoshi Matsuoka will define scientific AI priorities, whereas 30 breakout periods will vary from LLMs for most cancers analysis to bettering vitality effectivity in supercomputing. The end result guarantees to be half workshop, half technique summit, and an early have a look at how open, frontier‑scale AI could reshape science within the subsequent decade.
Agentic Tutorials and Hackathon
Organizers have framed this system round three simple targets: develop the contributor neighborhood via tutorials and a hackathon, supply a wider view in a sequence of plenary talks, and preview and create tasks throughout a slate of breakout periods.
For the primary day and a half, there shall be tutorials and hackathons to deliver extra arms on deck within the pursuits of creating frontier AI extra accessible for researchers. Operating in parallel with the tutorials, the hackathon will pair newcomers with skilled builders in groups of three. Every group will construct a easy agent inside an agentic framework, reminiscent of a DOI lookup bot or a light-weight internet search helper. This setup will guarantee a wholesome mixture of ability ranges that organizers hope will spark peer studying.
The give attention to brokers displays how agentic workflows have gotten normal in science via platforms reminiscent of Microsoft Uncover, OpenAI’s Deep Analysis, and Google DeepMind’s AI co-scientist. By the top of the session, individuals ought to depart with sensible expertise they’ll apply to their very own analysis.
Plenary Highlights: Leaders Map the Street Forward
The plenary periods will supply a bigger view of the priorities of the worldwide analysis computing neighborhood. Rick Stevens, affiliate laboratory director at Argonne and an AIwire Individual to Watch, will open the plenary sequence with “Reinventing Discovery: Accelerating Science within the Age of Synthetic Tremendous‑Intelligence.” Stevens will focus on how frontier‑scale fashions have moved past present as activity‑particular AI instruments to serving to form scientific questions and strategies. In his opening speak, he’ll hint how fast positive factors in speculation technology, simulation, and information interpretation now conflict with slower steps reminiscent of chemistry, fabrication, and discipline work. Borrowing from Amdahl’s Regulation, he’ll define the problem as a rethink of all the analysis workflow moderately than a single leap in compute. Stevens can even sort out sensible questions: how one can set targets that replicate human priorities, maintain outcomes reliable when artificial information is used, and determine which duties nonetheless require human experience. He plans to shut with concrete actions the neighborhood can take to navigate this new panorama collectively.
AI researchers like Rick Stevens, Satoshi Matsuoka, and Charlie Catlett shall be collaborating in TPC25. (Supply: TPC25)
Thierry Pellegrino of AWS will make the case for hybrid HPC, or linking on‑premises techniques with cloud capability, as the sensible path ahead. His speak will present how AI instruments, classical simulations, and early quantum assets can share one workflow, drawing transient examples from climate prediction, computational fluid dynamics, and molecular modeling which have sped up compute jobs with out extra spending.
Satoshi Matsuoka, director of RIKEN’s Middle for Computational Science, will define what nonetheless stands between giant language fashions and on a regular basis scientific work. Drawing on RIKEN’s AI4Science program and up to date runs on the Fugaku and Frontier supercomputers, he’ll focus on strategies for dealing with multi‑terabyte information, pairing neural surrogates with physics solvers, and spreading workflows throughout CPU, GPU, and early quantum {hardware}. Early ends in local weather modeling, 3‑D imaging, and supplies discovery present as much as 6.9x velocity‑ups, and Matsuoka will shut by flagging open questions round provenance and human oversight for petascale-and-beyond ML stacks.
Different plenaries will look at technical and moral issues as AI strikes deeper into science. Ian Foster of Argonne will define an “AI‑native discovery platform” that marries trillion‑parameter fashions to simulations, data graphs, and autonomous labs. Ricardo Baeza‑Yates of the Barcelona Supercomputing Middle will supply a survey of irresponsible AI within the type of bias and disinformation earlier than mapping the governance and regulatory instruments meant to curb it. Microsoft’s Preeth Chengappa will ask who holds company when AI brokers develop into ubiquitous, and what meaning for possession and management of outcomes. Karthik Duraisamy (College of Michigan) will define a proposed “lively inference” structure, a layered system that pairs fashions able to counterfactual reasoning with steady actual‑world validation, to shut three cussed divides in analysis AI: the abstraction hole, the reasoning hole, and the fact hole.
Collectively, plenary periods like these and the many others deliberate for TPC will assist attendees consider the successes and challenges of frontier‑scale AI in scientific apply.
Breakout Tracks: Turning Concepts into Working Teams in BoF Classes and Lightning Talks
(Supply: Shutterstock)
The third purpose of TPC25 is to drive targeted collaboration via afternoon breakout periods. There are 30 breakout slots unfold throughout six rooms and 5 time blocks. Most periods will observe a working‑group format: 90‑minute blocks that open with 4 to 6 lightning talks and depart time for targeted dialogue. Some tracks span a number of blocks, giving observe‑up periods room to refine concepts.
Birds‑of‑a‑Feather meetups will run in parallel. Subject tracks cowl each speedy analysis wants and long-term infrastructure questions. Area science goes deep within the Life Sciences lane, the place three periods look at AI for most cancers, agentic bio workflows, and basis fashions for biology. Lightning talks will focus on matters like peptide‑binder design, genome‑scale language fashions, and precision‑well being pipelines. Different periods will sort out vitality effectivity for AI at supercomputing websites and plans for a basis mannequin that represents {the electrical}‑grid distribution system. Organizers anticipate a number of of those casual meetups to mature into full TPC working teams as soon as the assembly concludes.
Why TPC25 Issues for AI‑Pushed Science
With sensible periods, plenary talks, and breakouts that would flip concepts into prototypes, TPC25 will present a transparent have a look at frontier AI for science and the folks guiding its future. Tutorials and the hackathon will assist attendees acquire new abilities, whereas plenary talks will define AI’s affect on analysis. Birds of a Feather conferences and lightning talks will give tasks the momentum they should advance. For researchers, system builders, and anybody following AI’s influence on science, the San Jose gathering shouldn’t be missed.
To see an agenda and be taught extra in regards to the Trillion Parameter Consortium and TPC25, go to https://tpc25.org and register at https://tpc25.org/register.
