/Tech3h ago

Meta's Lucas Beyer and Cody Blakeney argue modern plotting tools lack the precision of ggplot and Seaborn

Story Overview

Two AI researchers with deep experience at Meta and elsewhere are noting that today's plotting libraries speed up initial figure creation for papers yet still fall short on the fine control needed for clean, consistent results across variants and appendices.

619021.3K

#72

Original post

Cody Blakeney@code_star#1674inTech

I hate to say this, but I almost miss writing Seaborg or ggplot code.

Are my plots 10x easier to make now? Sure. Refactoring it easier? Absolutely.

But the satisfaction of having that super clean, elegant figure after you spent ages insuring that each variant of it in the paper and appendix was consist.

Man it felt good.

You would just show your lab mates, and yall would talk about it for 10-20 minutes. You would show the tricks you used, or maybe how you calculated the x / y axis for all the plots at once so plots could be directly compared for scale and meaningful interpreted.

Sure. Things are better now. Lord knows raw matplotlib was a pain in the ass. Yeah, definitely better now.

12:54 AM · Jun 23, 2026 · 1K Views

Workflow Friction

The quick-win phase leaves final tweaks painful

Modern approaches get rough drafts done fast for exploration, but reaching the last bit of polish, scale matching, and cross-figure consistency remains harder than with Seaborn or ggplot.

Open Question

Publication figures still demand manual craft

Collaborative review and attention to reproducibility details have not fully transferred to the newer tools, leaving an open gap for anyone preparing peer-review ready output.

Sentiment

Many users affirm the real satisfaction of manually crafting plots with Seaborn and ggplot rather than selecting from dropdowns.

Pos

100.0%

Neg

0.0%

2 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS468LIKES10REPLIES2

Lucas Beyer (bl16)@giffmana

@code_star Yep. But also, vibe-plotting gets you to 80% in 20% of the time, which is great for exploring results and presentation styles quickly.

But, so far, I've had quite a lot of trouble getting the last 20% to a really good plot.

Cody Blakeney@code_star

I hate to say this, but I almost miss writing Seaborg or ggplot code.

Are my plots 10x easier to make now? Sure. Refactoring it easier? Absolutely.

But the satisfaction of having that super clean, elegant figure after you spent ages insuring that each variant of it in the paper and appendix was consist.

Man it felt good.

Sure. Things are better now. Lord knows raw matplotlib was a pain in the ass. Yeah, definitely better now.

2h468100

snow@snowclipsed

@giffmana @code_star There are last mile problems everywhere for those with eyes to see

29m321

Zach Mueller@TheZachMueller

@code_star Me with manim rn. Tried going agentic and now I’m doing cursor cmd+k loop + my own coding like it’s 2023

20m41

Cody Blakeney@code_star

@giffmana ASI is here when it can oneshot this

32m15

Ozar@ozarliquid

@code_star ngl theres a diff between building it by hand and selecting it from a dropdown

that satisfaction is real though

1h3

KIKI@0xkivaro

@code_star there was something satisfying about earning a great figure the hard way 👏

41m2

Taro Bushidō@techietaro

@code_star Training from scratch vs fine-tuning, same dynamic. Faster now but you lose that granular understanding. The struggle built intuition. Worth it though.