Sharon Machlis

Author of Practical R for Mass Communication & Journalism
Still ❤️ R after all these years #RStats
Can't imagine life without writing
Having fun with #GenAI
Also learning #ASL and #Python
Newly retired (voluntarily!) after a long career as a tech journalist. Eagerly exploring my next chapter.
Passionate about vibrant aging
Other interests: #Photography #DigitialDarkroom #Running #Bicycling #Crochet
Joined Mastodon at Fosstodon Oct 2022

Public Key

npub1rsnd0ksgq9hfvquvmcxs3vm8pv375nfru7vq98gwwnm9svhcp0ss0julvp

NIP-05 Address

smach_at_masto.machlis.com@mostr.pub

Profile Code

nprofile1qqspcfkhmgyqzm5kqwxdurggkdnskgl2f5370xqzn588fajcxtuqhcgpzemhxue69uhhyetvv9ujumn0wd68ytnzv9hxgqgkwaehxw309aex2mrp0yh8qunfd4skctnwv46qn27903

Publishing to

relay.nostr.band

relay.primal.net

Author Public Key

npub1rsnd0ksgq9hfvquvmcxs3vm8pv375nfru7vq98gwwnm9svhcp0ss0julvp

Show more details

Last Notes

2025-04-15 01:02:06

Although OpenAI's GPT 4.1 model family isn't available in ChatGPT, you can try GPT 4.1 free in the Windsurf code editor until April 21.
https://windsurf.com/editor
As with Cursor (and VS Code), Windsurf isn't my favorite environment for writing and running R code - but it works. And the #GenAI integration is nice.
Hoping Positron gets more a more 'native' #LLM experience. That's been teased a couple of times 🤞

2025-02-12 22:44:30

#OpenAI roadmap: "We will next ship GPT-4.5, the model we called Orion internally, as our last non-chain-of-thought model.... We want AI to 'just work' for you; we realize how complicated our model and product offerings have gotten.
"After that, a top goal for us is to unify o-series models and GPT-series models by creating systems that can use all our tools, know when to think for a long time or not, and generally be useful for a very wide range of tasks." Sam Altman 1/2
#GenAI #OpenAI #LLMs

2025-01-15 02:49:40

I just added langfuse tracing & analysis to a #GenAI hobby project I'm working on, and IT WAS SO EASY. A few lines of Python code + a free account, and I have SO much data on what's going on - including #LLM costs, which I care about if I make it public someday.
https://langfuse.com

2024-12-28 19:31:57

I tried out the new {gander} #RStats 📦 that creates a chatbot within either RStudio or Positron to help with your code. What's nice is you can choose your context: None, current file, or file + neighbors. Same #LLM options as {ellmer}.
https://simonpcouch.github.io/gander/
By @nprofile…5s3m #GenAI
It wasn't immediately obvious (to me) how to change models, but these worked:
options(.gander_fn = "chat_claude") # default
options(.gander_fn = "chat_openai")
options(.gander_args = list(model = "gpt-4o"))
Etc.
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/732/100/487/884/233/original/e09797389cef809c.jpg
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/732/102/147/111/521/original/f75b774c3adad76d.jpg
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/732/103/333/856/855/original/f0c84f3a5ac4cf95.jpg

2024-12-24 19:42:23

Claude can now "directly analyze Excel files to pull data, evaluate formulas, recreate a formatted spreadsheet as an Artifact, and more," Anthropic says. While it could analyze CSV files before, checking Excel formulas is new.
Claude Sonnet 3.5 is probably my favorite all-around #LLM . I like its default response style, it think it's usually best at summarizing documents, and several experts I trust say it's best at writing R code.
#GenAI

2024-12-16 01:11:04

Is it just me as an inexperienced Python user? Or are Python package dependency conflicts a real pain even when using virtual environments? I feel like I've had more of these conflicts in Python this weekend working on a GenAI hobby project than in the last 10 years of using R. . . .

2024-12-04 13:04:21

Batch process #LLM requests in #R ”by looping across rows of a data frame column.” The {batchLLM} #RStats 📦 uses “data frames and column rows as [LLM] input and a new column with the text completions as the output.” Supports OpenAI, Claude & Gemini. Includes a Shiny add-in.
“My work on a complex classification problem inspired me to create this tool,” says author Dylan Pieper. On CRAN.
#GenAI
https://github.com/dylanpieper/batchLLM
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/594/683/336/244/799/original/21d3bf8556ca1846.jpeg

2024-11-23 14:12:10

The {tidyprompt} #RStats 📦 *still under development* is designed for prompting your #LLMs “the tidy way.” It includes function calling, so you “can define your own R functions and give the LLM the ability to call them and process their output.”
By Luka Koning & Tjark Van de Merwe
https://tjarkvandemerwe.github.io/tidyprompt/
@nprofile…25un #GenAI #LLM
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/532/674/313/772/150/original/13496e82677d8976.jpeg

2024-11-23 13:31:56

The {scico} #RStats 📦 features “39 different palettes, all of which are perceptually uniform and colourblind safe. An overview can be had with the scico_palette_show() function”. By @nprofile…5up2
https://github.com/thomasp85/scico
#R #DataViz #DataVis @nprofile…25un
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/532/516/017/572/239/original/fae65e55e91680bf.png

2024-11-20 21:54:01

From Dr. Albert Rapp (via LinkedIn) "It took me waaay too long to learn about str_view(). 🫨 It's one of the best helper functions to understand regular expressions. It shows you exactly where the matches to your regular expression are with <brackets>."
Me: How did I not know about this until now???
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/517/503/495/061/471/original/5d93e16f737a9ae1.jpeg

2024-11-20 03:17:05

In a study comparing ChatGPT, doctors using ChatGPT, and doctors without an #LLM, ChatGPT “scored an average of 90% when diagnosing a medical condition from a case report and explaining its reasoning,” The NY Times reports. Doctors using ChatGPT averaged just 76% - in part because some didn’t like the chatbot's diagnoses even when correct - and doctors alone 74%.
https://www.nytimes.com/2024/11/17/health/chatgpt-ai-doctors-diagnosis.html?unlocked_article_code=1.bU4.qFm-.hw7BFPq4F-qg&smid=nytcore-ios-share&referringSource=articleShare
#GenAI

2024-11-16 18:22:32

From Kyle Walker: "Explore a choropleth map of work-from-home patterns by PUMA from the 1-year American Community Survey for 2023.
"Make it yourself in #rstats https://github.com/walkerke/map-challenge-2024/blob/main/day-16-choropleth/day-16-script.R
"Browse the live map here: https://walker-data.com/map-challenge-2024/day-16-choropleth/ "
#RSpatial #R #GIS @nprofile…25un
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/494/021/135/177/028/original/839d0d5cccd4ace1.jpg

2024-11-13 12:20:48

“What tasks can an #LLM do well?” turns out to be a complicated question. This article looks at whether “emergence” - when #LLMs seem to suddenly develop new skills - is truly a thing:
https://cacm.acm.org/news/is-it-possible-to-truly-understand-performance-in-llms/
#GenAI

2024-11-09 13:08:38

Researchers fine-tuned a Mistral large language model to detect racist covenants in real estate deeds. They also worked with Santa Clara County, CA to develop responsible usage, “including legal review and the creation of a historical registry.”
The research team says this was massively more efficient than manual review and cheaper than using a closed model. They've released the #LLM “to assist the hundreds of jurisdictions engaged in similar efforts.”
https://reglab.github.io/racialcovenants/
#GenAI

2024-10-25 19:51:56

💯agree with Stuart Stevens on this: By choosing not to endorse a candidate, the LA Times and Washington Post "are violating rule 1 of Timothy Snyder's warnings 'On Tyranny.'
"1. Do not obey in advance. Most of the power of authoritarianism is freely given. In times like these, individuals think ahead about what a more repressive government will want, and then offer themselves without being asked. A citizen who adapts in this way is teaching power what it can do."
#USPolitics #USPol

2024-10-24 15:42:27

The goal of the {elmer} #Rstats 📦"is to provide a user friendly wrapper over the most common llm providers. Major design goals include support for streaming and making it easy to register and call R functions." On GitHub. By @npub1re9…dtna
Currently supports OpenAI, Anthropic, and Google.
https://hadley.github.io/elmer/
#GenAI #LLMs #LLM

2024-10-24 01:59:37

On a local trail today (my photo + a Topaz Iimpressionist filter)
#Photography #Autumn
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/359/919/266/258/449/original/8b383e13d6401a22.jpg

2024-10-18 18:19:36

Gorgeous couple of autumn days in southern New England
#photography #autumn
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/329/775/109/906/579/original/01f48fd9f45168d9.jpg
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/329/784/763/019/744/original/8c021947f0dfa8b5.jpg
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/329/799/398/598/631/original/6aa926a04728bd29.jpg

2024-09-29 23:21:09

There's a CustomGPT specifically for R and RStudio -- that is, ChatGPT with prompts specifically focused on R. It should be usable for free ChatGPT accounts, not only paid ones. By Jose A Fernandez Calvo
https://chatgpt.com/g/g-iedyANQtM-r-and-r-studio-tutor
#RStats #R #ChatGPT #GenAI
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/223/400/599/628/309/original/d8a59428ecaf8ddb.jpg

2024-09-18 18:05:30

People can't make informed decisions about how much risk they're willing to accept *if they don't know what the risks are.*
Even mild cases of Covid-19 can damage the brain and cardiovascular system, increase likelihood of diabetes, and result in long Covid. Do people know this and don't care? Or they're told “it's mild now” and don't know?
If they care, how are they informed when rates are high or indoor spaces are poorly ventilated and unsafe?
#CovidIsNotOver https://aus.social/@DenisCOVIDinfoguy/113156356866261651

2024-09-13 19:40:58

For economic policy nerds, from Mark Cuban on That Other App: Trump Tariff vs VP Harris's corporate tax rate (COGS = cost of goods sold)
#USPolitics #USPol @npub1aa9…umwm
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/131/943/165/445/491/original/8b206fa0c8ca3bdd.jpeg

2024-08-31 19:50:12

It's grey and cloudy out now, but we got in a few hours of nice weather in Hull, MA, including lunch and a walk along Nantasket Beach. Happy Labor Day weekend to all in the US celebrating!
#Photography
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/058/360/019/057/313/original/26d6574dd2634a69.jpg
https://cdn.masto.host/mastomachliscom/media_attachments/files/113/058/360/044/385/789/original/14489766c4966e73.jpg

2024-08-26 21:50:14

- reply

@npub103v…93fg Isn't mean(x, na.rm = TRUE) accurate R code?

2024-08-14 16:39:33

"Think Stats" 3rd edition in progress by #PositConf24 keynoter Allen Downey, is available free online at https://allendowney.github.io/ThinkStats/ in Jupyter notebooks. Code in #Python

2024-08-13 16:38:08

Future of PDFs: Quarto can now generate PDFs without Latex! It does this by using a new markup-based typesetting system under development called typst. "You can produce really beautiful PDFs" Charlotte Wickham tells #PositConf2024
#QuartoPub
https://quarto.org/docs/output-formats/typst.html

2024-08-12 21:32:36

To deal with FOMO not being in Seattle for the PositConf2024 data science conference (I'll be attending remotely starting tomorrow), I had a lovely day in Boston today with a friend, including a ferry ride across Boston Harbor.
#photography
https://cdn.masto.host/mastomachliscom/media_attachments/files/112/951/183/370/080/980/original/993ab52a512a6e3e.jpg

2024-08-05 15:54:19

Morning walk, Cochituate Rail Trail, Natick, Mass.
#Photography
https://cdn.masto.host/mastomachliscom/media_attachments/files/112/910/219/040/582/117/original/3d451befd01f1265.jpg

2024-07-04 02:11:06

“We often dismiss skills that are not societally valued by pretending they are not skills. Or, sure, maybe they're skills, but they're mysterious and ineffable!
“You reinforce this idea each time you say ‘soft skills’ instead of working to enumerate them. Perhaps you mean ‘interpersonal skills,’ ‘leadership skills,’ or ‘communication skills.’ These are all technical skills, and they all have names.” - Sasha Laundy 🔥🔥🔥
⬇️⬇️⬇️ https://social.jvns.ca/@b0rk/112722455355582173

2024-05-22 16:40:53

I cannot overstress how envious I am of democracies that have election campaigns of 6 weeks instead of 2 years

2024-05-22 12:05:55

#Climate threat not often discussed:
“Pound for pound, gallon for gallon, hour-for-hour, the two-stroke gas powered engines in leaf blowers and similar equipment are vastly the dirtiest and most polluting kind of machinery still in legal use.
“According to the California Air Resources Board (CARB), the two-stroke leaf blowers and similar equipment in the state produce more ozone pollution than all of California’s tens of millions of cars, combined.”
- James Fallows
https://fallows.substack.com/p/gas-powered-leaf-blowers-the-end

2024-05-14 22:07:49

I'm so lucky to have a botanical garden less than a 10-minute bike ride away and work from home. Lunch hour today:
#photography
https://cdn.masto.host/mastomachliscom/media_attachments/files/112/441/712/006/768/942/original/211ed9dee5061ce0.jpg
https://cdn.masto.host/mastomachliscom/media_attachments/files/112/441/714/491/333/246/original/403db5e0924843cf.jpg
https://cdn.masto.host/mastomachliscom/media_attachments/files/112/441/716/506/168/099/original/f4f12a7bbf28b2e5.jpg

2024-04-20 12:06:05

I'm baffled by people who've already decided #GenerativeAI isn't worthwhile because <insert current limitation here>. It's like dismissing the Internet because who'd want to use a 300 baud modem that ties up their landline.
We're still in the #LLM early days. We don’t know how #GenAI will improve.

2024-04-20 12:06:05

2024-03-17 20:08:09

Update: I paid $20 for a month of Claude Pro to use Anthropic's GPT-4-class model, uploaded City Council meeting minutes, and used a modified version of Anthropic's “Cite your sources” suggested prompt for my queries. Responses to questions about what issues specific City Councilors discussed at the meeting were surprisingly good - and included source references. Worth further investigation.
#LLM #LLMs #GenAI #GenerativeAI #AI https://masto.machlis.com/@smach/112108143932301141

2023-11-26 12:43:49

Massachusetts does not collect data on police traffic stops *unless a ticket is issued.* This makes it impossible to analyze data about racial bias in stops where no driver wrongdoing was found.
“Why the fight to curb racial profiling via traffic stop data keeps failing in Mass”
https://www.metrowestdailynews.com/story/news/state/2023/11/17/the-mass-legislature-repeatedly-rejected-anti-racial-profiling-bills/70516496007/

2023-11-20 20:32:14

Hey U.S. folks: Starting today, you can order 4 more free at-home Covid-19 tests, even if you ordered some earlier this fall. Stock up now for pre-holiday-gathering testing!
https://www.covid.gov/tests
#Covid19 #CovidIsNotOver

2023-10-30 18:38:41

Tidy time series & forecasting in R - AFRICAST workshop
Slides: https://workshop.f4sg.org/africast/
GitHub: https://github.com/Forecasting-for-Social-Good/workshop-f4sg-africa
"AFRICAST is a new development of the Democratising Forecasting initiative https://www.f4sg.org/services/democratising-forecasting/ from the Forecasting for Social Good (F4SG) at the International Institute of Forecasters, which aims to make forecasting best practices available to a broader community, particularly in low and lower-middle income countries." - Bahman Rostami-Tabar
#rstats @npub15y6…mhcy
https://cdn.masto.host/mastomachliscom/media_attachments/files/111/325/412/844/684/861/original/8d31883b620da9db.jpeg

2023-10-18 19:01:59

- reply

@npub1efv…4jx2 If you're an Excel user, Excel on MacOS is somewhat terrible.
(And yes, I know one can use other software and export .xlsx files, I do that with R all the time. But I have a friend who works in Finance who won't get a Mac just because of Excel.)

2023-08-18 03:31:01

- reply

@npub1f6a…pnql Adding embeddings to llm would be great if possible!! 🔥🔥

2023-08-13 21:13:39

Little Island park along the Hudson River, New York City
#NewYork #NYC #photography #summer
https://cdn.masto.host/mastomachliscom/media_attachments/files/110/884/363/443/133/691/original/24d48adf222f90b8.jpg

2023-07-20 10:28:30

6 tips on how you - and everyone else - can see more interesting posts on Mastodon.
For ex., did you know you're probably not seeing all public posts with hashtags when you search? There are ways to see more!
https://www.infoworld.com/article/3701292/6-ways-to-find-better-content-on-mastodon.html
#Mastodon #Fediverse #MastodonMigration #FediTips #InfoWorld
@npub1094…2wum @npub1g7t…n6x7 @npub17wy…p8qc

2023-07-03 13:23:05

“Intro to Data Science with R: Unlocking the power of R for public health data science”
New PDF to accompany free video courses by University of Geneva's GRAPH Network. Project supported by WHO.
https://drive.google.com/file/d/1Rw-hdMtlsGRbCF0yA52yQRcLVQ0aVHv9/view
Video courses: https://thegraphcourses.org/courses-portal/
R Foundations, Data Wrangling with dplyr, #DataViz with #ggplot2, Introduction to EpiGraphHub, and Intro to Geospatial Visualization with R
#rstats #rspatial #dplyr @npub15y6…mhcy
https://cdn.masto.host/mastomachliscom/media_attachments/files/110/650/339/300/704/799/original/b27b039984a8d7c5.jpeg