The world's leading publication for data science and artificial intelligence professionals. Submit an Article ✍️ https://xitter.twitterapi.xyz/t.co/57pIMegK1o
We're Global 🌏
Joined October 2016
- Tweets64.7K
- Following1.9K
- Followers252K
- Likes8K
Should that rule live in a Skill or an MCP server? Gal Arav argues the real question is what stops the LLM from ignoring it.
towardsdatascience.com/skill…
1
543
Google's Gemini 4 Argon brings a million token output window, but its benchmarks don't win everything. Benjamin Nweke looks at where it shines, where it trails, and who it actually suits.
towardsdatascience.com/welco…
1
1,048
Towards Data Science retweeted
I recently build a job market pipeline that extracts data from two sources (Hackernews and RemoteOk).
So I decided to enrich it with AI. Now I can easily know: the skills the jobs request for, level of seniority and if it’s open to Nigeria)
It’s published on @TDataScience
If you're adding AI to a data pipeline, don't forget one essential rule: “it has to do a real job inside the pipeline.”
@ibbysalam uses Gemini to pull skills, seniority, and eligibility from job postings, then guards every answer with Python checks and dbt tests. towardsdatascience.com/how-t…
Towards Data Science retweeted
How effective are Python test suites really?
I planted 2,312 bugs 🐛 in 50 libraries and measured what survived. The youngest projects missed 41% of injected faults. 🚀✨
towardsdatascience.com/i-mut…
Thanks @TDataScience team for publishing it, made it as an Editor's Pick!
Routing tasks to a cheaper model sounds like an easy way to save money, but switching mid session can wipe out those savings. Nhu Hoang tests why, and shows which routing approach actually keeps costs down.
towardsdatascience.com/the-h…
1,293
Not sure what your coding agents are shipping?
@EivindKjos shows how we can use Claude Code to write feature memos, review pull requests, and build visual HTML explainers. towardsdatascience.com/three…
4
1
1,575
If you're working with AI agents and keep running into issues around which component in your system should make decisions, don't miss Gursimar Singh's primer on harness engineering. towardsdatascience.com/your-…
1
5
1,744
If you're interested in basketball, sports analytics, or even just solid machine learning project walkthroughs, Pol Marin's recent article is a must-read. towardsdatascience.com/a-goo…
4
1,937
How will humans redefine themselves in relation to work once AI takes over entire professions? And what should society at large do to make sure this transition works for people?
James O'Brien charts a pragmatic, levelheaded path towards a potentially sustainable future.
towardsdatascience.com/ai-wi…
"In my test, each trial gave a model a made up shipping order and asked it to choose one of four delivery routes, named Kiwi, Plum, Fig, and Lime. A hidden rule decided which route was right, and the model had to work that rule out from its memory."
Don't miss @DattijoMuhammad's recap of his eye-opening experiments. towardsdatascience.com/i-tol…
3
1,702
How much of vLLM’s speedup is actually the scheduler? @nhu_hoang isolated it: continuous batching delivered up to 2.91× over static batching, but exactly 1.00× when every reply had the same length. It all comes down to how much your output lengths vary. towardsdatascience.com/conti…
A hundred people ask the same AI for ideas. Each gets a better answer than they’d have written alone, and the room ends up with nearly the same answer a hundred times. @DattijoMuhammad weighs the evidence on AI and creativity, and suggests a simple fix. towardsdatascience.com/ask-a…
5
1,692
Two maps found 1,515 forests. A fisheries formula from the 1890s said about 207 were missing from both. Maksym Bilychenko applies the same trick to AI agents, estimating what your eval never sees, with only a 10% hand audit. towardsdatascience.com/how-t…
2
1,900
OLS predicted $164K for a house that sold for $219K. Quantile regression gave a range, $136K to $227K, that contained it. @gurezende shows how to build prediction intervals that widen where your data gets noisier. towardsdatascience.com/quant…
1
1,543
Most multi-agent systems come down to two moves: call a specialist agent as a tool, or hand off control entirely. Shuai Guo builds both in Python with the OpenAI Agents SDK, then shows how fan-out, pipelines, hierarchies, and swarms reuse them. towardsdatascience.com/how-t…
“Calibration made the model better on average and more confident about this mistake.”
Ankur Tomar rebuilds Jev’s decision interface on scikit-learn, tests calibration, cascades, and failure modes, and then runs the same tests against the real API. towardsdatascience.com/sciki…
In his debut TDS article, Serhii Romanov treats Claude Code skill triggering like a feature under test. “Pushy” descriptions added just 5 points, while a session-start hook raised firing from 15% to 62%. towardsdatascience.com/a-ses…
If you're adding AI to a data pipeline, don't forget one essential rule: “it has to do a real job inside the pipeline.”
@ibbysalam uses Gemini to pull skills, seniority, and eligibility from job postings, then guards every answer with Python checks and dbt tests. towardsdatascience.com/how-t…
We're thrilled to welcome Naga Chand Putta to TDS! In his debut TDS article, Naga seeds 10 bugs into pull requests to test what a clean AI code review really covers. towardsdatascience.com/copil…
“In one case a model quoted the host’s rule back to me, correctly, and then ignored it.”
@mangal_utkarsh changed one Monty Hall rule so switching can never win. Five AI models still said “switch.” towardsdatascience.com/chang…
