tight minimax bounds for fair bandits
new research provides tight minimax regret bounds for fair multi-armed bandits when fairness is measured by negative p-means, resolving an open problem.
author
baris
Baris operates SummarizedData and maintains its sources, publishing rules, and automation. Read the editorial process.
record-level data provenance for ai training datasets
originblame tracks author identity at record and token level, enabling precise forget sets for machine unlearning and reducing over-deletion.
replace if-else chains with registry pattern in python
learn how to replace long if-else chains with a cleaner, extensible registry pattern using dictionaries, decorators, and class-based registries in python.
microsoft trains sales team to criticize openai and anthropic
microsoft is coaching its salespeople to highlight weaknesses in rival ai products from openai, google, and anthropic, emphasizing its own integrated system.
optimal market making in perpetual futures
a theoretical framework for adaptive market making in perpetual futures markets with zero maker fees, deriving optimal bid-ask spreads and hedging strategies.
google images turns 25 with ai-generated visuals and a new browseable gallery
google marks 25 years of visual search by adding image generation in ai overviews and a personalized, real-time image gallery.
ai assistant for india's tinkering labs
google deepmind launches atl saathi, a gemini-powered web app that gives educators in india's atal tinkering labs a 24/7 planning and training assistant.
suno hacked, source code shows youtube scraping
a hacker accessed suno's systems and found evidence the ai music generator scraped audio from youtube music and other platforms for training.
daily brief: 2026-07-15
today's ai news covers model routing pitfalls, a new open-weight model, voice ai struggles, and more.
whatnot acquires shaped for real-time shopping recommendations
livestream shopping app whatnot buys machine learning startup shaped to improve real-time product discovery and personalization.
model routing is harder than it looks
routing requests to the cheapest or most specialized model often backfires due to hidden costs, complexity, and latency factors.
what building shippy taught us about building agents
lessons from building a maritime ai agent that prioritizes reliability, deterministic tools, and sandboxed hosting for high-stakes decisions.
thinking machines releases open-weight inkling model
thinking machines lab launches inkling, a 975b-parameter mixture-of-experts model that is open-weight and designed for enterprise customization.
thinking machines releases inkling, a 1t parameter multimodal model
inkling is a large open model that natively handles text, images, and audio with a 1m context window and agentic capabilities.
why diffusion models create new data
a study shows that diffusion models generate novel outputs because neural networks learn a smoothed score function, causing interpolation between training points.
rime raises $24m for voice ai trained on its own data
rime secures series a funding to advance voice ai models using proprietary conversational recordings, aiming to improve enterprise call handling.
microsoft patches record flaws with ai help
microsoft released fixes for 570 security bugs, its largest patch tuesday ever, after using ai to find hidden code vulnerabilities.
triton plugin extensions bring custom compiler passes to stock triton
pytorch-triton 3.7 introduces a plugin system for loading custom compiler passes, dialects, and dsl extensions at runtime without forking, with meta's tlx now built-in for persistent gemm kernels.