Tokenless

Tokenless

If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com

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
If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com

If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com

Ramakrishnan Lokanathan argues that AI evals are broken because they are overcomplicated, and lays out a practical framework for building them for high-stakes applications: define the most ambitious version of success, test every step of a workflow, combine automated checks with human review and adversarial tests, guard against regressions, and turn real user failures into tests. He also introduces eSlams, an open-source package where models compete in games like chess, Go, and poker as a window into how they reason.

Your AI Agent Wants Your Data. Here's How to Lock It Down - Jeff Smith, Oracle

Your AI Agent Wants Your Data. Here's How to Lock It Down - Jeff Smith, Oracle

Oracle's Jeff Smith explains how to let agents query enterprise databases safely through an MCP server, covering reports exposed as vetted tools, the risks of natural language to SQL, performance tuning on read-only standbys, schema annotations, and deep data security enforced at the database layer.

Everyone Has the Same AI Models. Here's What Sets Agents Apart - Richmond Alake, Oracle

Everyone Has the Same AI Models. Here's What Sets Agents Apart - Richmond Alake, Oracle

Richmond Alake argues that after the control plane, agent architectures need an intelligence plane covering what an agent knows, remembers, is allowed to see, and how it improves. He draws on the WWII Midway codebreaking story to separate information from intelligence, then demos a governed agent using an ontology and semantic layer that answers correctly about 99% of the time with far fewer tokens than a naive agent. The talk also covers Oracle's agent memory package, deep data security, private inference, and fine-tuned decision models for memory routing.

Technology

View All
Why Founders Are Betting Big on Factories, Mines, and Energy

Why Founders Are Betting Big on Factories, Mines, and Energy

Erin Price-Wright and Connor Love unpack the AD50 and the rebuilding of America's industrial base, covering how defense demand is financing manufacturing capacity for commercial markets, how AI is reshaping factory economics, how a new generation of founders applies lessons from SpaceX, Tesla, Anduril, and Palantir to mining, energy, and manufacturing, and why private capital is now willing to fund physical infrastructure before government demand is guaranteed.

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.

Recent Post

Q-learning with Flow-Matching Policies

Q-learning with Flow-Matching Policies

This talk explores methods for optimizing expressive, multi-modal policies, such as those based on flow-matching, with off-policy reinforcement learning. The speaker presents two novel algorithms, FQ-RL and CAM, designed to overcome the instability of backpropagation through multi-step generative models, enabling effective online self-improvement and adaptation for robotic manipulation tasks.

Wavefunction Flows: Efficient Quantum Simulation of Continuous Flow Models

Wavefunction Flows: Efficient Quantum Simulation of Continuous Flow Models

Continuous flow models map naturally to a Schrödinger equation, the fundamental equation of quantum mechanics. This discovery proves that a trained generative model can be efficiently simulated on a future quantum computer, enabling a new, more powerful type of access to its learned distribution for tasks like Monte Carlo estimation and structure discovery.

Where the Score Lives: What Wavelets Reveal About Diffusion Models

Where the Score Lives: What Wavelets Reveal About Diffusion Models

This talk explores the paradox of why diffusion models generalize rather than memorize. It introduces an analytically tractable, wavelet-based parameterization of the score function, allowing for an interpretable analysis of how architectural biases (like locality) and data statistics interact to influence denoising performance and generalization.

How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL

How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL

Cursor's Federico Cassano and Fireworks' Dmytro Dzhulgakov detail their collaboration on Composer 2, a specialized foundation model for software engineering. They discuss their top-down training strategy, the infrastructure challenges of large-scale distributed Reinforcement Learning on sparse models, and how model specialization achieves frontier performance with superior efficiency.

End-to-End Foundation Models for the Energy Industry — with Jazmia Henry

End-to-End Foundation Models for the Energy Industry — with Jazmia Henry

Jazmia Henry details the end-to-end process of building specialized foundation models for the energy industry. She covers the four key stages from data curation of unstructured, handwritten documents to optimizing inference, and introduces her Grounded Continuous Evaluation (GCE) framework to combat reward hacking in reinforcement learning.

The Four Types of Memory Every AI Agent Needs

The Four Types of Memory Every AI Agent Needs

AI agents utilize four distinct types of memory, analogous to human cognition, to move beyond simple chatbot responses. This summary explores the CoALA framework, detailing working, semantic, procedural, and episodic memory and how they enable agents to learn, recall skills, and leverage past experiences.

← PreviousNext →

Stay In The Loop! Subscribe to Our Newsletter.

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