Desmond Choy (@Norest) header
Desmond Choy (@Norest) profile picture

Builder. Student. Dad. Banner image by @liamwong.

Singapore
Joined March 2009
Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)Photo by Desmond Choy (@Norest)
Encountered some buggy behaviour with G3.1 on @antigravity. From initial testing, it's capable and managed to proposd more elegant solutions vs O4.6, but "common sense" is lacking e.g. committing files that it didn't touch. (O4.6 would never do that)
1
90
One aspect where GPT-5 shines (and I don't see much praise for) is its integration with Jupyter Notebooks. It understands and edits Jupyter Notebooks much better than Sonnet or Gemini. Editing notebooks with Codex VSCode Extension is *chef's kiss*
114
Some thoughts after using GPT-5 exclusively in @AmpCode for the last week. Initially, there were occasional flashes of brilliance, but the model was unreliable, as it would hang occasionally (router issue). I also saw instances of it doing tool calls in Russian.
2
2
574
But of late it's been very very useable and I haven't missed Sonnet 4. It's slower than Sonnet 4, takes more time to think, but it's also less opinionated and follows instructions very well.
1
1
40
Is it just me or is @AmpCode 's Oracle really slow all of a sudden? Usually would take 1-2 mins but now I observed really slow thinking.
3
3
1,128
"Feeling the AGI" with @AmpCode and playwright MCP. Giving the agent "vision" is a game changer.
1
16
658
Halfway through @HamelHusain & @sh_reya’s #AIEvals course and it’s already making me rethink how I approach my LLM projects and conduct error analysis. The course has a very high signal-to-noise ratio. Highly recommended!
2
7
2,164
After using @AmpCode for a week and burning through $100 in credits and counting (thanks @GeoffreyHuntley), here are some thoughts: By design, it seems amp takes a very surgical approach in looking for things within files, which helps in conserving total token count.
1
14
933
The to-do list it creates is an interesting feature. Good UX (letting user know what's planned) but I'm not sold if it actually Claude plan better. It occasionally makes horrible mistakes (adding one col to SQLite as the solution, but deleting two for absolutely no reason)
1
563
Thanks to @eugeneyan for the idea & inspiration: Can buying life insurance be less painful/boring? Instead of manually completing forms, complete your risk assessment through an AI voice-powered underwriter instead. Colab: github.com/DesmondChoy/ai_un… nitter.cf/eugeneyan/status/17708…
1
71
Is there a way to tune #tidymodels algorithms on the cloud? I have a large dataset (2m rows, 30 columns) and would like to speed up tuning of hyperparameters. Have played around with parallel processing in #r and the finetune package, but still takes quite a while.
5
Amazing resource for any #ggplot2 users.
🔥🔥 MAJOR UPDATE 🔥🔥 The extended "#ggplot2 Tutorial for Beautiful Plotting in #Rstats" is online 🥳 A ton of new examples such as adding annotations, playing with after_scale, new chart types, modifying legends, interactive #dataviz and many more! 🔗 cedricscherer.netlify.app/20…
4
Lediona Nishani, Data Scientist at Veeva Systems nominated for the #WomenTech Global Awards 2020! Share to raise awareness! #womenintech #womeninstem #womenwhocode #WTGA2020 @womentechnet womentech.net/nominee/Ledion…
1
1
What makes a cup of coffee memorable? Using the #tidymodels framework, I explore a recent #TidyTuesday data set containing coffee quality data and train three models to predict coffee rating scores. desmondchoy.netlify.app/2020… #rstats
2
10
2
78
This project showcases data cleaning, EDA, data visualization, imputing missing values using k-NN, feature engineering, three models (LASSO, Random Forest, XGBoost), hyperparameter tuning, checking for overfitting, and variable importance comparison.
1
5
#Singapore #GE2020 For anyone interested in comparing how Singapore's political parties differ across various issues (but don't wish to read the manifesto), here's an interactive table to conveniently search and compare topics that matter to you: bit.ly/3f3sPa8 #rstats
6
4
#rstats blog post: 🦵⚽️scored some EDA & visualization milestone goals 🥅using a #kaggle #FIFA20 player data set with the help of ridgeline plots, a network graph and unsupervised PCA with #tidymodels desmondchoy.netlify.app/2020…
2
4
I explored the relationship between share prices of four internet companies and sentiment harvested using #tidytext mining from quarterly earnings call transcripts. Hint: It's complicated. desmondchoy.netlify.app/2020… #rstats
A first stab at using #tidytext mining on hip-hop lyrics + #tidymodels to predict if a song is old-school (released <2000) or modern (>=2000). Extremely grateful to @juliasilge @drob for their amazing work: tidytextmining.com/ desmondchoy.netlify.app/2020…
5
Replying to @pbaumgartner
@pbaumgartner Your blogdown and disqus tutorials are invaluable and have been incredibly helpful - Danke! notes.peter-baumgartner.net/…
Having trouble narrowing down a specific discipline from the hundreds of free Springer e-books? Here are some word clouds highlighting popular (and potentially esoteric) sub-categories. Also blogged about it here: desmondchoy.netlify.app/2020… #rstats #tidyverse
Springer has released over 400 science books downloadable as pdf for free as part of 'Free Textbooks and Library Link special issue during Covid 19 package'. Lots of stats books and some ecology ones as well You can search and download here: link.springer.com/search/pag…
1
6
1
6