Sign in

Simon P. Couch

@simonpcouch.com
3.9K followers 214 following 386 posts

he/him - writing statistical software at Posit, PBC (née RStudio)🥑 simonpcouch.com, @simonpcouch elsewhere

PostsRepliesMedia
Simon P. Couch @simonpcouch.com · 16/09/2026
FYI Posit AI Pass subscribers; we wound down the instance serving Gemma 4 today. If you were a regular user, you might give GLM 5.3 Flash a try; it's vision capable, quite a bit larger and newer (and thus more capable!), and served at ~half the price that Gemma 4 was per-token!
071
Simon P. Couch @simonpcouch.com · 16/09/2026
It was such a highlight for me, too! @k-bott.bsky.social
010
Reposted by Simon P. Couch
Garrick Aden-Buie @grrrck.xyz · 15/09/2026
Wicked psyched to share that shinychat v0.5.0 #RStats and v0.7.1 #Python are now available! Check out all the rad new features on the Posit Open Source blog. #PositConf2026 opensource.posit.co/blog/2026-09...
opensource.posit.co
Complete chat applications in shinychat: R 0.5.0 and Python 0.7.1
shinychat v0.5.0 for R and v0.7.1 for Python make it easier to build complete, conversation-centered Shiny chat applications with history, editing, branching, greetings, suggestions, citations, tool d...
0203
Simon P. Couch @simonpcouch.com · 15/09/2026
😆😆
010
Reposted by Simon P. Couch
Kelly Bodwin @kellybodwin.com · 15/09/2026
Every time @simonpcouch.com makes a great point in his #positconf2026 Keynote, his earring sparkles extra brightly in the stage lights. 🌟
Simon Couch on stage, next to a projected slide showing an article title from Science journal warning about software errors.
1182
Simon P. Couch @simonpcouch.com · 15/09/2026
We're excited to share commons, an R and Python package that helps data scientists build trustworthy data agents for their organizations! Read more on the @posit.co open source blog: opensource.posit.co/blog/2026-09...
A screenshot of the blog post announcing commons. The hero image shows the hex sticker, a common kingfisher on a park bench.
14311
Simon P. Couch @simonpcouch.com · 14/09/2026
In this edition of My Coworkers Are Awesome, @hadley.nz had a real-life version of the blob on the chores sticker made!
Me, happily holding a plushie of a small blob holding a clipboard.The package logo, a small yellow blob happily holding a clipboard.
0411
Simon P. Couch @simonpcouch.com · 14/09/2026
So excited to listen to your talk!
020
Reposted by Simon P. Couch
Jon Harmon (he/him/his) @jonthegeek.com · 13/09/2026
#PositConf2026 (via the #RPharmaSummit2026 pre-game) has already been worthwhile for introducing me to Kimi K3 in Posit AI Pass! So far it's good, fast, and relatively cheap. This is what I get for skipping any of @simonpcouch.com 's blog posts! opensource.posit.co/blog/2026-08...
opensource.posit.co
Kimi K3 and GLM 5.2 are now in Posit AI
A new batch of open weights models in Posit AI show impressive capabilities at a fraction of the price.
131
Simon P. Couch @simonpcouch.com · 13/09/2026
What a lineup!
020
Reposted by Simon P. Couch
Simon Willison @simonwillison.net · 12/09/2026
"For a while, I must admit, it looked as if software developer roles like mine were done for. [...] But our industry is slowly realizing that making truly cutting-edge software still requires humans to think and work together, to maximize their skill sets and to practice their respective crafts."
512521
Simon P. Couch @simonpcouch.com · 11/09/2026
I saw our packages were hanging out in `precheck` together. :)
110
Simon P. Couch @simonpcouch.com · 11/09/2026
I AM SO EXCITED FOR POSIT CONF
2223
Simon P. Couch @simonpcouch.com · 11/09/2026
Will do :)
030
Simon P. Couch @simonpcouch.com · 11/09/2026
So stoked to see all of you!
130
Reposted by Simon P. Couch
Grace @gracekind.net · 10/09/2026
Crossposting to the swarm artifactory instance
2651
Simon P. Couch @simonpcouch.com · 09/09/2026
RUN ML MODELS DIRECTLY IN YOUR DATABASE
0156
Simon P. Couch @simonpcouch.com · 09/09/2026
this is thanks to @ivelasq3.bsky.social for reminding me!!! ty ty!
010
Simon P. Couch @simonpcouch.com · 08/09/2026
posit.co/conference is next week! So stoked! We got an order in for a Houston limited edition (ruby red grapefuit!) #rstats stacks sticker just in time. :)
Five hexagonal stickers for the 'stacks' R package from tidymodels.org, each featuring an illustration of a stack of pancakes with syrup and different fruit toppings. The word 'stacks' is arranged in a descending staircase pattern on each hex. The top-left 'Original' version has a light blue background with a steel blue border and blueberry topping. The top-right 'posit::conf(2026)' version has a burnt orange background with a dark brown border and grapefruit slices. The bottom row shows three more conference editions: 'posit::conf(2025)' in cream with a copper border and peach slices, 'posit::conf(2024)' in muted purple with a dark navy border and blackberries, and 'posit::conf(2023)' in soft pink with a rose-red border and strawberries.
1183
Simon P. Couch @simonpcouch.com · 07/09/2026
"I have tried (almost) nothing and it isn't working"🤣
031
Reposted by Simon P. Couch
Posit @posit.co · 07/09/2026
Can't make it to Houston? Virtual passes for posit::conf(2026) start at $99, with academic and nonprofit rates at $49 and a needs-based option at $0. Every keynote and session livestreamed, September 14–16. conf.posit.co
1257
Simon P. Couch @simonpcouch.com · 05/09/2026
🫶
000
Simon P. Couch @simonpcouch.com · 04/09/2026
New on the @posit.co blog: In my experience, fine tuning LLMs is more difficult and more expensive than I've seen it made out to be. In most cases, I'd recommend focusing your efforts on experimenting with a bunch of prompts and existing models. opensource.posit.co/blog/2026-09...
opensource.posit.co
AI Newsletter: You probably don't want to fine-tune
You're usually better off doing plain old prompt engineering.
2225
Simon P. Couch @simonpcouch.com · 03/09/2026
5 months ago, models in the ~30B A3B range started to be able to do simple agentic coding tasks. I recently wondered how 8B (or even 4B!) models would fare: simonpcouch.com/blog/2026-09...
Agentic coding reliability for five recent local models. Granite 4.2 8B succeeded in 80 percent of runs and Ornith 1.5 9B in 60 percent. Granite 4.2 3B, LFM 2.5 2.6B, and Qwen 3.8 4B Distill all scored zero.
061
Reposted by Simon P. Couch
Garrick Aden-Buie @grrrck.xyz · 02/09/2026
GLM 5.3 Flash is really fun. Super zippy and super cheap, and it’s great at getting things done. Pair it with a smart model or a smart human and you can get places fast!
0102
Simon P. Couch @simonpcouch.com · 02/09/2026
To use the model, open up Posit Assistant in RStudio or Positron and update when prompted. Then, find it in the model selector. :)
010
Simon P. Couch @simonpcouch.com · 02/09/2026
We just shipped GLM 5.3 and GLM 5.3 Flash in Posit AI Pass! GLM 5.3 Flash is now the cheapest model available via the subscription (half the price of Gemma 4 26B A4B per-token!) and I've been so, so impressed with it. GLM 5.3 is Ox Alpha, the anonymous model that was popping off on OpenRouter.
Table comparing AI model costs per token, sorted from cheapest to most expensive. Columns: Model, Lab, and Relative Cost Per-Token. GLM 5.3 Flash (Z.ai) is cheapest at 0.05x; Gemma 4 26B (Google, going away soon) at 0.1x; Claude Haiku 4.5 (Anthropic, going away soon) at 0.33x; GLM 5.3 (Z.ai) and GLM 5.2 (Z.ai, going away soon) both at 0.38x; Claude Sonnet 5 (Anthropic) at 0.67x; Claude Sonnet 4.6 (Anthropic) at 1x as the baseline; Kimi K3 (Moonshot AI) at 1x; and Claude Opus (Anthropic) as the most expensive at 1.67x.
1142
Simon P. Couch @simonpcouch.com · 29/08/2026
One is short enough to fit in a bsky post :) bsky.app/profile/smac... The btw MCP server includes all sorts of tools, but if you're interested in a more limited, self-contained example, this post might be helpful: opensource.posit.co/blog/2026-07...
opensource.posit.co
mcptools 1.0.0
The first major release of mcptools, an R SDK for the Model Context Protocol, is now on CRAN.
120
Simon P. Couch @simonpcouch.com · 28/08/2026
This post demoes out both directions of the http protocol: opensource.posit.co/blog/2026-07...
opensource.posit.co
mcptools 1.0.0
The first major release of mcptools, an R SDK for the Model Context Protocol, is now on CRAN.
041
Simon P. Couch @simonpcouch.com · 28/08/2026
As of mcptools 1.0.0, you should be good to go on that now! Feel free to holler if you ever end up having a moment to give it another whir and it doesn't support your needs.
110
Simon P. Couch @simonpcouch.com · 28/08/2026
Ah, I'm seeing now that you referenced this in a different reply.
000
Simon P. Couch @simonpcouch.com · 28/08/2026
`btw::btw_mcp_server()` is an mcptools MCP server that ships several R/ellmer-based tools :) posit-dev.github.io/btw/referenc...
posit-dev.github.io
Start a Model Context Protocol server with btw tools — mcp
btw_mcp_server() starts an MCP server with tools from btw_tools(), which can provide MCP clients like Claude Desktop or Claude Code with additional context. The function will block the R process it's ...
220
Simon P. Couch @simonpcouch.com · 19/08/2026
hooooly smokes
020
Simon P. Couch @simonpcouch.com · 17/08/2026
How do you choose an LLM? A survey of the current model landscape, including results from the vibes-from-Sara-and-Simon eval (/s) New on the @posit.co open source blog: opensource.posit.co/blog/2026-08...
opensource.posit.co
AI Newsletter: How to choose a model
Which model is 'best'? A survey of the current model landscape.
1103
Simon P. Couch @simonpcouch.com · 17/08/2026
let's see the guitar!👀
100
Simon P. Couch @simonpcouch.com · 10/08/2026
I'll be there. Stoked to see you!
020
Simon P. Couch @simonpcouch.com · 10/08/2026
I've been really impressed with Kimi K3 and GLM 5.2 the last couple weeks. I'm excited to share that we've made the models available in Posit AI with strong privacy guarantees and a steep discount compared to the Claude models currently available in the service! opensource.posit.co/blog/2026-08...
opensource.posit.co
Kimi K3 and GLM 5.2 are now in Posit AI
A new batch of open weights models in Posit AI show impressive capabilities at a fraction of the price.
0335
Simon P. Couch @simonpcouch.com · 05/08/2026
i was looking up song lyrics and
Screenshot of a Google search for "don't mean to wake you up." The AI Overview box responds: "You are not disturbing anything. I am ready to help you right now," followed by a "How to Proceed" section suggesting the user share a question, topic, or task, and asking "What would you like to work on today?" Below the AI Overview is a "Show more" expander, then a YouTube search result for Ken Yates' song "Don't Mean To Wake You (Acoustic)."
081
Simon P. Couch @simonpcouch.com · 03/08/2026
Data analysis is often a branching and nonlinear process. We just shipped a feature in Posit Assistant to help with this; use /eda-log to keep track of loose ends in your analysis. @posit.co #databs #rstats opensource.posit.co/blog/2026-07...
opensource.posit.co
AI Newsletter: EDA log in Posit Assistant
A higher-level view of your data analysis conversations.
0254
Simon P. Couch @simonpcouch.com · 01/08/2026
ooooooomg this sticker
130
Simon P. Couch @simonpcouch.com · 29/07/2026
this got old so fast
030
Reposted by Simon P. Couch
Max kuhn @topepo.bsky.social · 29/07/2026
We're happy to announce our new #rstats package: lorax. If you fit tree-based models, lorax enables conversion to the party package's format (for making nice tree diagrams) and has methods to compute which features were used, the number of terminal nodes, etc. opensource.posit.co/blog/2026-07...
opensource.posit.co
Introducing lorax: Speaking for the Tree-Based Models
lorax is a new R package that characterizes fitted tree- and rule-based models: extract their decision rules, see which predictors they actually use, and convert trees for plotting with partykit.
26112
Reposted by Simon P. Couch
Carlos Scheidegger @cscheid.net · 28/07/2026
I am begging you to talk to other human beings directly, and not through a Claude layer. This goes doubly when I ask you, directly, to talk to me not through a Claude layer. Please.
2474
Simon P. Couch @simonpcouch.com · 27/07/2026
A new release of mcptools, an #rstats package implementing the Model Context Protocol, is now on CRAN! It's a patch release with several security-oriented fixes. github.com/posit-dev/mc...
github.com
Release mcptools 1.0.1 · posit-dev/mcptools
This release includes several security-oriented fixes, in addition to a couple quality of life improvements for multi-user and multi-session workspaces: The server now chooses its R session at too...
0173
Reposted by Simon P. Couch
Hadley Wickham @hadley.nz · 24/07/2026
This week on my substack: a write up of my talk "y code when ai" (my thoughts on why knowing how to code is still relevant), along with a brief description of how I turned a video of my talk into this post. Read it at open.substack.com/pub/tidydesi... !
open.substack.com
y code when ai?
A talk on AI and coding
2477
Simon P. Couch @simonpcouch.com · 24/07/2026
actually LOLed
120
Reposted by Simon P. Couch
Jeremy Allen @jeremy-data.bsky.social · 24/07/2026
It’s Friday. Do a breakthrough.
3173
Reposted by Simon P. Couch
Tomasz Kalinowski @t-kalinowski.bsky.social · 23/07/2026
I'm happy to share `ir`, a new command-line tool for running portable R scripts and Quarto documents. It's inspired by my two favorite parts of `uv`: self-describing scripts and tools you can run without installing first. opensource.posit.co/blog/2026-07...
opensource.posit.co
ir 0.1.0: self-describing R scripts and Quarto documents
`ir` is a new command-line tool for running portable R scripts and rendering Quarto documents whose package requirements, and optional R selection, live inside the file itself.
47521
Simon P. Couch @simonpcouch.com · 21/07/2026
Just ran today's Gemini 3.6 Flash release through #rstats bluffbench2. In the ballpark of Gemini 3.5 Flash on performance, slightly (5-10%) cheaper. More on the eval: posit-dev.github.io/bluffbench2/
A bar plot showing scores for several frontier models. The two leaders, Gemini 3.5 Flash and Claude Fable 5, score in the mid-high teens. Gemini 3.6 Flash is around 10%. Models from OpenAI cluster at the bottom, never eclipsing 10%.
070
Simon P. Couch @simonpcouch.com · 21/07/2026
One of the more interesting findings to me from this eval is that prematurely adding modeled results (like #rstats geom_smooth()) to plots drastically reduces the chances that models will catch the issue in the plot.
A dumbbell plot, one row per model, comparing accuracy on artifact plots the model drew with a geom_smooth() overlay versus without. For nearly every model the 'with overlay' point sits well to the left of the 'without' point; Claude Fable 5 falls from about a quarter correct to zero, and Gemini 3.5 Flash from about a quarter to under a tenth.
170