Tootfinder

Google releases a new Gemini Robotics On-Device model with an SDK and says the vision language action model can adapt to new tasks in 50 to 100 demonstrations (Ryan Whitwam/Ars Technica)
https://arstechnica.com/google/2025/06/google-relea…

Google’s new robotics AI can run without the cloud and still tie your shoes
Google’s Carolina Parada says Gemini has enabled huge robotics breakthroughs, like the new on-device AI.

Robotics & AI, a lot of progress is being made.
#AI

Google’s new robotics AI can run without the cloud and still tie your shoes
Google’s Carolina Parada says Gemini has enabled huge robotics breakthroughs, like the new on-device AI.

The 4th Dimension for Scaling Model Size
Ruike Zhu, Hanwen Zhang, Tianyu Shi, Chi Wang, Tianyi Zhou, Zengyi Qin
https://arxiv.org/abs/2506.18233 https://…

The 4th Dimension for Scaling Model Size
Scaling the size of large language models typically involves three dimensions: depth, width, and the number of parameters. In this work, we explore a fourth dimension, virtual logical depth (VLD), which increases the effective algorithmic depth without changing the overall parameter count by reusing parameters within the model. Although parameter reuse is not a new concept, its potential and characteristics in model scaling have not been thoroughly studied. Through carefully designed controlled…

Geometry of Chord Intertwiner, Multiple Shocks and Switchback in Double-Scaled SYK
Sergio E. Aguilar-Gutierrez, Jiuci Xu
https://arxiv.org/abs/2506.19013 h…

Geometry of Chord Intertwiner, Multiple Shocks and Switchback in Double-Scaled SYK
We revisit the bulk Hilbert space interpretation of chords in the double-scaled SYK (DSSYK) model and introduce a notion of intertwiner that constructs bulk states from states with fixed boundary conditions. This leads to an isometric map that factorizes the one-particle bulk Hilbert space into a tensor product of two boundary Hilbert spaces without particle insertion. The map enables a systematic derivation of a family of correlation functions with arbitrary finite amount of matter insertions,…

Trans-dimensional Hamiltonian model selection and parameter estimation from sparse, noisy data
Abigail N. Poteshman, Jiwon Yun, Tim H. Taminiau, Giulia Galli
https://arxiv.org/abs/2506.18802

Trans-dimensional Hamiltonian model selection and parameter estimation from sparse, noisy data
High-throughput characterization often requires estimating parameters and model dimension from experimental data of limited quantity and quality. Such data may result in an ill-posed inverse problem, where multiple sets of parameters and model dimensions are consistent with available data. This ill-posed regime may render traditional machine learning and deterministic methods unreliable or intractable, particularly in high-dimensional, nonlinear, and mixed continuous and discrete parameter spac…

Stratified Hyperbolicity of the moduli stack of stable minimal models, II: Big Picard Theorem and the stratified Brody hyperbolicity
Junchao Shentu
https://arxiv.org/abs/2506.19218

Stratified Hyperbolicity of the moduli stack of stable minimal models, II: Big Picard Theorem and the stratified Brody hyperbolicity
This is the second paper on the global geometry of Birkar's moduli of stable minimal models (e.g., the KSBA moduli stack). We introduces a birationally admissible stratification of the Deligne-Mumford stack of stable minimal models, such that the universal family over each stratum admits a simple normal crossing log birational model. The main result of this paper is to show that for each stratum $S$, the pair $(\overline{S},\partial S)$ satisfies the Big Picard theorem. In particular, we show t…

How Robust is Model Editing after Fine-Tuning? An Empirical Study on Text-to-Image Diffusion Models
Feng He, Zhenyang Liu, Marco Valentino, Zhixue Zhao
https://arxiv.org/abs/2506.18428

How Robust is Model Editing after Fine-Tuning? An Empirical Study on Text-to-Image Diffusion Models
Model editing offers a low-cost technique to inject or correct a particular behavior in a pre-trained model without extensive retraining, supporting applications such as factual correction and bias mitigation. Despite this common practice, it remains unknown whether edits persist after fine-tuning or whether they are inadvertently reversed. This question has fundamental practical implications. For example, if fine-tuning removes prior edits, it could serve as a defence mechanism against hidden …

A US judge rules that Meta's use of books to train AI is protected by fair use but says his opinion is more a reflection of the plaintiffs' poor arguments (Isaiah Poritz/Bloomberg Law)
https://news.bloomberglaw.com/legal-op

Meta Beats Copyright Suit From Authors Over AI Training on Books
Meta Platforms Inc. escaped a first-of-its-kind copyright lawsuit from a group of authors who alleged the tech giant hoovered up millions of copyrighted books without permission to train its generative AI model called Llama.

Stratified Hyperbolicity of the moduli stack of stable minimal models, I
Junchao Shentu
https://arxiv.org/abs/2506.17537 https://arxi…

Stratified Hyperbolicity of the moduli stack of stable minimal models, I
In this paper, we introduce a birationally admissible stratification on the Deligne-Mumford stack of stable minimal models (e.g., the KSBA moduli stack), such that the universal family over each stratum admits a simple normal crossing log birational model. We further demonstrate that each stratum is hyperbolic in the sense that every schematic generically finite covering of any closed substack is of logarithmic general type. This provides a partial answer to C.Birkar's question regarding the gl…

Salesforce launches Agentforce 3 with an observability tool called Command Center and MCP support, and says 8,000 customers have signed up to deploy Agentforce (Larry Dignan/Constellation Research)
https://www.constellationr.com/blog-news/i

Salesforce launches Agentforce 3, Command Center for visibility
Salesforce launched Agentforce 3, which features Agentforce Command Center, support for Model Context Protocol (MCP), and updated Atlas architecture to speed up reasoning and performance. The rollout maintains Salesforce's Agentforce cadence, which includes updates every few months as the company learns from enterprise use cases. In addition, Salesforce has added more than 30 partners to AgentExchange including AWS, Box, Cisco, Google Cloud, IBM and payments players such as PayPal and Stripe.