Tokenless

Tokenless

How Anthropic Merges 30,000 Code Changes a Day - Manish Kapur, Sonar & Addy Osmani, Anthropic

Interactive discovery

Explore the topic map

Follow the connections between themes, people, and ideas across the Tokenless archive in an interactive topic modeling map.

Open the map Browse posts

Machine Learning

View All
Judge a Transaction by the Company It Keeps — Somya Rai & MD Sharique, EXL

Judge a Transaction by the Company It Keeps — Somya Rai & MD Sharique, EXL

EXL developed an advanced fraud detection system that augments XGBoost with Graph Neural Network (GNN) embeddings. Traditional tabular models struggle to capture fraud signals residing in transaction neighborhoods. The system's innovative design includes meticulous sampling strategies, disjoint monthly graph construction for leakage control, and a GraphSAGE architecture that aggregates mean, max, and standard deviation of neighbor features to detect subtle fraud patterns. Emphasizing explainability for regulatory compliance and analyst trust, it also features a highly optimized serving architecture achieving 3ms P99 latency, pivoting to edge prediction to reduce cache footprint from 60GB to 2-3GB. This led to a 27-33% relative lift in true positive rates and a projected $65 million incremental fraud capture.

From 15% to 90% GPU Utilization: Fix the Data Pipeline, Not the Model

From 15% to 90% GPU Utilization: Fix the Data Pipeline, Not the Model

Tarun Sunkaraneni from Amazon AGI shows how a multimodal SFT pipeline that spent about 85% of its time waiting on data was brought to roughly 90% GPU utilization. He walks through concurrency with asyncio and Ray actors, prefetching, zero-copy transfer via Ray's object store, and a network-card bottleneck that appears only at scale, and argues that the hardest part of training throughput is often keeping GPUs fed rather than optimizing the model.

Small Models, Big Results: Training a Finance Agent for Under $500 — Charles Dickens, Snorkel AI

Small Models, Big Results: Training a Finance Agent for Under $500 — Charles Dickens, Snorkel AI

Snorkel AI and UC Berkeley's Sky Computing Lab show that a 4B Qwen3 model, trained with reinforcement learning for under $500 using the rLLM framework, beats its 235B sibling on expert-validated financial question answering built from SEC 10-K filings. The talk covers the FinQA dataset construction and verification, common failure modes of frontier models, transfer to multi-table and general tool-use tasks, and ablations showing that simple data and binary rewards outperform more elaborate setups, since the real bottleneck is disciplined tool use rather than reasoning depth.

Artificial Intelligence

View All
Why a One-Line Change Can Break Your Whole Stack - Cole Gannaway, Qodo

Why a One-Line Change Can Break Your Whole Stack - Cole Gannaway, Qodo

placeholder

How Anthropic Merges 30,000 Code Changes a Day - Manish Kapur, Sonar & Addy Osmani, Anthropic

How Anthropic Merges 30,000 Code Changes a Day - Manish Kapur, Sonar & Addy Osmani, Anthropic

Manish Kapur of Sonar and Addy Osmani of Anthropic discuss why coding agents are opening pull requests faster than teams can review them, how light and dark software factories differ, how Anthropic scaled to around 30,000 merged changes a day with Claude doing first-pass review backed by verification evidence, new State of Code survey data, and why every loop in the factory needs independent, multi-layered verification.

AI That Hunts Logic Bugs Before Attackers Do - Taylor Luttrell-Williams, Sonar

AI That Hunts Logic Bugs Before Attackers Do - Taylor Luttrell-Williams, Sonar

Sonar's Taylor Luttrell-Williams introduces SonarQube Hunter Agent, an AI security agent that finds intent-level flaws such as broken access control, business logic gaps, and authentication holes that pattern-based SAST cannot detect. The talk covers its multi-agent pipeline with adversarial validation, Sonar's join to the OpenAI Daybreak Cyber Partner Program, and a live demo that surfaced 46 issues in a Python app, including a self-promotion to admin and an MFA bypass.

Technology

View All
Mark Cuban’s Plan to Make Healthcare Affordable

Mark Cuban’s Plan to Make Healthcare Affordable

Mark Cuban and Julie Yoo discuss how transparency, direct contracting, and AI could reshape healthcare economics, from Cost Plus Drugs and modular pod manufacturing to personalized medicine, the future of physicians, and advice for founders who want to use AI to learn rather than to stop thinking.

Exploits Now Beat Patches: Open Source Security in the AI Era - Patrick Smyth, Chainguard

Exploits Now Beat Patches: Open Source Security in the AI Era - Patrick Smyth, Chainguard

Patrick Smyth of Chainguard examines how frontier AI has compressed the time between vulnerability disclosure and exploitation, leaving 30-day remediation cycles obsolete. The talk covers AI-assisted attacks, waves of supply chain malware, four attack surfaces spanning malware, known CVEs, zero days, and side doors like CI/CD, developer laptops, and agent skills, and practical defenses including cooldown periods, faster CVE remediation, and signing artifacts with SLSA and Sigstore.

Can Your AI Heat My House?

Can Your AI Heat My House?

A wide-ranging conversation on how AI inference is pushing cloud infrastructure, data centers, and hardware supply chains against physical limits, covering the rise of alt clouds, rack density and cooling, hardware lifecycles and circularity, talent shortages, heat reuse for communities, and what engineering leaders can do about resource-aware software.

Recent Post

SOTA Generative Media Panel — Dumitru Erhan, Shane Gu & Nicole Brichtova, Google DeepMind

SOTA Generative Media Panel — Dumitru Erhan, Shane Gu & Nicole Brichtova, Google DeepMind

This detailed discussion with Google AI researchers highlights the latest generative video models (NanoBanana 2 Lite, Gemini Omni Flash APIs) and their applications, including video editing, education, and localization. The conversation critically examines whether language is an adequate intermediate representation for AI, especially for sensory experiences like audio and taste, where human vocabulary is limited. It exposes the unreliability of human preferences in evaluating AI-generated content, citing the "Instagram filter effect" and "wedding ring" reward hacking as examples. The researchers emphasize the importance of high-quality, real-world data and the critical role of forward-deployed engineers in bridging the gap between cutting-edge research and practical applications.

AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)

AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)

Tara Seshan, who leads product for Codex and ChatGPT Work at OpenAI, explores how AI is shifting knowledge work from rowing to steering, why ambition and human judgment are becoming key differentiators, how to build for models two to three months ahead, and how writing as thinking still matters. The conversation also covers OpenAI's founder-led culture, the design of Work mode, and practical uses like Sites and visualize.

Why Does AI Need Access to the Web?

Why Does AI Need Access to the Web?

This video addresses the critical limitation of pre-trained LLMs: their static knowledge in a dynamic world, leading to hallucinations and costly mistakes for AI agents. It introduces the concept of 'knowledge layers' empowered by 'web data infrastructure' to provide AI with live, fresh, reliable, and properly formatted web data, thereby ensuring grounded and timely responses and eliminating the 'garbage in, garbage out' problem for real-world AI applications.

The Signal Layer: What to Build When Anything Can Be Built — Lena Hall, Akamai

The Signal Layer: What to Build When Anything Can Be Built — Lena Hall, Akamai

Lena Hall explores the paradox of AI's abundance: while it grants immense leverage, it also makes average work valueless as competitors gain the same capabilities. She argues that AI, acting as a "convergence machine," produces identical answers from common knowledge. The true challenge shifts from building to "pointing"—deciding which problems are worth solving. Hall emphasizes that "taste" is trainable by AI, but genuine "judgment" about future events or unique customer relationships is not. She identifies critical points of signal distortion in startups and large organizations and advocates for engineering a "thin signal layer" to preserve unique intent. Ultimately, in a world of infinite alternatives, the scarcest and most valuable commodity is trust, which cannot be automated or graded.

Agentic Sites: Building Hyper Personalized Websites — Carlos Sanchez, Adobe

Agentic Sites: Building Hyper Personalized Websites — Carlos Sanchez, Adobe

Carlos Sanchez from Adobe presents "Agility Sites," a novel approach to building hyper-personalized websites in real-time. By dynamically generating specific content blocks and grounding responses in the site's corpus, these sites achieve an "audience of one" experience. Key innovations include rapid LLM inference (1.1s average latency with Cerebras), continuous model evaluation for speed and accuracy, and leveraging user intent to deliver tailored content, as demonstrated by an on-the-fly camping coffee machine page and an AI conference comparison site.

Agents Are Where Microservices Were in 2015 — Roberto Milev & Uday Kanagala, Navan

Agents Are Where Microservices Were in 2015 — Roberto Milev & Uday Kanagala, Navan

Roberto Milev and Uday Kanagala from Navan discuss the challenges and emerging best practices in building and operating AI agents, drawing parallels with microservices. They highlight the shift to stateful runtimes, innovative memory and context management using 'skills,' and crucial solutions for observability and testing non-deterministic systems. A central theme is the critical need for robust authorization and guardrails, given the blurring line of agentic action, and a preference for mastering single-agent loops before scaling to complex orchestrations. They conclude by outlining the solved and unsolved problems, with cost and debugging remaining significant hurdles.

← PreviousNext →

Stay In The Loop! Subscribe to Our Newsletter.

Get updates straight to your inbox. No spam, just useful content.