Towards Data Science (@TDataScience) header
Towards Data Science (@TDataScience) profile picture

Towards Data Science

@TDataScience

The world's leading publication for data science and artificial intelligence professionals. Submit an Article ✍️ https://xitter.gq/t.co/57pIMegK1o

 We're Global 🌏
towardsdatascience.com
Joined October 2016
  • Tweets64.7K
  • Following1.9K
  • Followers252K
1.7KMedia
Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)Photo by Towards Data Science (@TDataScience)
Every model switch means rebuilding context the new model has never seen. Nhu Hoang breaks down the hidden cache cost behind LLM routing. towardsdatascience.com/the-h…
The Hidden Cost of LLM Routing: Why Switching Models Costs More

The Hidden Cost of LLM Routing: Why Switching Models Costs More

I tested LLM routing to understand why switching models mid-session can erase the savings promised by cheaper inference

towardsdatascience.com
1
6
961
What can you hand a model that writes up to a million tokens back in one go? Benjamin Nweke walks through Gemini 4 Argon, from setup to pricing. towardsdatascience.com/welco…
Welcome to Google Gemini 4 Argon

Welcome to Google Gemini 4 Argon

A million-token output window changes what you can hand a model in one go, and the benchmarks and the bill show who it actually suits.

towardsdatascience.com
3
2,052
A Skill can only advise an LLM, while code can actually refuse. Gal Arav maps twelve common formats by whether they inform or enforce, with lessons from shipping both. towardsdatascience.com/skill…
Skill or MCP Server? Ask What Stops the LLM From Disobeying

Skill or MCP Server? Ask What Stops the LLM From Disobeying

Towards Spec-Driven Test Automation, Part 4

towardsdatascience.com
1
2
8
2,139
Let a decision model pick the next action and an LLM write the text. Shuai Guo's hands on guide shows how this hybrid setup finished the same task in a fraction of the time. towardsdatascience.com/decis…
Decision Models in Action: Build a Hybrid LLM Workflow

Decision Models in Action: Build a Hybrid LLM Workflow

A hands-on guide to building browser automation by combining decision models with LLMs.

towardsdatascience.com
2
1
15
2,267
What's the real ROI of your AI initiatives? Weiwei Hu presents a 12-KPI framework to allow you to measure it accurately. towardsdatascience.com/how-t…
Hours Saved Aren't Dollars Saved. A 12-KPI Framework for Measuring Real AI ROI

Hours Saved Aren't Dollars Saved. A 12-KPI Framework for Measuring Real AI ROI

Includes a worked example with 50 users and $110,000 in first-year cost, plus baselines, targets, and ownership rules for every AI initiative.

towardsdatascience.com
1
2
5
1,669
Thinking about switching to a different Python environment? Sara A. Metwalli weighs the pros and cons of uv, venv, and pip. towardsdatascience.com/how-t…
You Don't Need to Abandon venv, but uv Makes Managing Environments Much Easier

You Don't Need to Abandon venv, but uv Makes Managing Environments Much Easier

A side-by-side walkthrough of venv plus pip and uv init, uv add, uv run, and uv sync, with a lockfile that recreates the project on another machine.

towardsdatascience.com
2
7
1,818
Ready to take your agent-building skills to the next level? @taupirho's new guide shows how to create a multi-tool agent with Ollama and Python. towardsdatascience.com/build…
Cap Your Agent Loop at Eight Steps Before One Bad Call Becomes Hundreds

Cap Your Agent Loop at Eight Steps Before One Bad Call Becomes Hundreds

AI Agents for Beginners, Part 2: The agent picks between three database tools, uses each result to choose its next call, and stops at an eight-turn circuit-breaker if it gets stuck.

towardsdatascience.com
2
6
1,822
Does your agent need persistent memory? It probably does. Mahnoor Javed explains how to give it precisely that in a step-by-step tutorial. towardsdatascience.com/give-…
An AI Agent That Remembers Your Name Across Restarts, Built with One JSON File

An AI Agent That Remembers Your Name Across Restarts, Built with One JSON File

Three Python classes, an OpenRouter key, and gpt-4.1-mini give the agent a calculator, a stub web search, and conversation recall that survives restarts.

towardsdatascience.com
1
2
9
2,173
We're thrilled to welcome Val Ye to TDS! Don't miss her excellent debut article, which outlines the process for sharing customized reports to 100+ clients with one semantic model. towardsdatascience.com/how-t…
Build Time Fell 70% When 100+ Clients Shared One Semantic Model

Build Time Fell 70% When 100+ Clients Shared One Semantic Model

Configuration-driven reporting: one semantic model, one metric catalog, and a few lines of configuration per audience

towardsdatascience.com
2
3
1,927
"Finding a localized gas plume within hundreds of gigabytes of hyperspectral data, however, is a massive needle-in-a-haystack problem." @previtus joins TDS with an insightful recap of his NASA JPL project: the Daily Methane Digest, which slashes alert latencies from months to ~24 hours and achieves zero-shot generalization to unmodeled gases. towardsdatascience.com/how-w…
Trained on Methane, Our Model Spotted Carbon Monoxide Plumes from Space

Trained on Methane, Our Model Spotted Carbon Monoxide Plumes from Space

We trained one model on methane. Without retraining, it also maps carbon monoxide, ammonia and nitrogen dioxide plumes, including the first automated CO detections from NASA’s EMIT.

towardsdatascience.com
2
3
1,730
What shape with the internet take in the future, and can businesses prepare for it? Rodrigo Almeida unpacks the stakes behind the rise of AI summaries, bot paywalls, and more. towardsdatascience.com/how-g…
Google AI Summaries Halve Clicks to Publishers. Bot Paywalls Charge AI Instead

Google AI Summaries Halve Clicks to Publishers. Bot Paywalls Charge AI Instead

The internet of the future will look much different than the one today. Businesses are already transitioning.

towardsdatascience.com
1
1
6
1,712
What's behind LLMs' two-hop curse? @mangal_utkarsh explains why a model might possess two discrete facts and still struggle with a question that relies on both. towardsdatascience.com/why-a…
I Tested a One-Line Fix for the Two-Hop Curse. It Only Half Worked.

I Tested a One-Line Fix for the Two-Hop Curse. It Only Half Worked.

A tiny GPT memorized every fact I gave it, then couldn’t answer a single question that needed two of them. I tested the one-line fix from a recent paper, and it only half worked.

towardsdatascience.com
4
9
2,022
Towards Data Science retweeted
Thomas Reid (@taupirho) profile picture
Thomas Reid@taupirhoNov 6
The second of my 3 part article series AI Agents for Beginners is now on @TDataScience. Read how to use multiple tools, incorporate memory and cap model turns to prevent runaway loops track.pstmrk.it/3s/towardsda…
1
1
533
Still counting stock with a clipboard? Samir Saci shows how warehouse operators can just say their counts out loud while an n8n workflow updates the inventory sheet for them. towardsdatascience.com/ai-in…
AI Inventory Cycle Count Tool

AI Inventory Cycle Count Tool

Warehouse operators speak their counts — "location J9, five pieces" — and an n8n workflow with speech-to-text updates the inventory sheet hands-free.

towardsdatascience.com
1
5
1,577
Abdullahi Dattijo put agentic RAG head to head with plain search across 2,000 arXiv papers. Here's what the extra searches gained, what they missed, and why it may not be worth it. towardsdatascience.com/does-…
Does Agentic RAG Actually Retrieve Better Evidence? I Tested It on 298 Questions

Does Agentic RAG Actually Retrieve Better Evidence? I Tested It on 298 Questions

I let an AI agent search again whenever it chose. It found some papers plain search had missed, lost some plain search had found, and gained too little to trust.

towardsdatascience.com
3
13
2,692
New to harnesses and harness engineering? Don't miss Sam Black's one-stop resource for learning about this emerging — and increasingly essential — field within agentic AI. towardsdatascience.com/what-…
What Is Harness Engineering? A Hands-On Introduction

What Is Harness Engineering? A Hands-On Introduction

Learn three essential primitives of harness engineering to future-proof your career

towardsdatascience.com
2
1,616
"A new model lands at the top of a leaderboard with 71.8%. The runner-up has 70.0%. The table makes that look like a result." @mangal_utkarsh explains why we should often take these rankings with a healthy dose of skepticism. towardsdatascience.com/is-th…
Is the #1 AI Model Really #1?

Is the #1 AI Model Really #1?

On a 2,000-question test, my top two models couldn't be told apart at the usual 5% level. On a 200-question version, a different model would have taken first place about three times in ten.

towardsdatascience.com
1
1
3
1,756
Lagged-outcome regression or difference-in-differences? Ryan O'Sullivan unpacks the benefits and shortcomings of the two approaches in the context of product analytics. towardsdatascience.com/diffe…
Difference-in-Differences vs Lagged-Outcome Regression

Difference-in-Differences vs Lagged-Outcome Regression

A simulation study and a pre-period check for choosing the right design for self-selected treatments in product analytics.

towardsdatascience.com
6
2,309
In her recent deep dive, Theophano Mitsa zoomed in on different approaches to categorizing data, using vision-language model CLIP to compare labels' and embeddings' performance. towardsdatascience.com/when-…
When Labels Say No, But Embeddings Say Yes

When Labels Say No, But Embeddings Say Yes

Why traditional retrieval metrics miss meaningful similarities

towardsdatascience.com
1
5
1,749
What are we actually measuring when we model credit risk? Junior Jumbong's new deep dive unpacks the metrics and components that make a difference. towardsdatascience.com/credi…
Credit Risk: What Are We Actually Measuring?

Credit Risk: What Are We Actually Measuring?

Let’s start with a loan and work out what PD, LGD, and EAD tell us about losses and bank capital.

towardsdatascience.com
2
1,542
Load more