Microblog

I really loved Euruko 2022's closing keynote by @andycroll. It has an important message and some great storytelling. I'm planning on reading more about Steve Shirley during my winter holidays. https://www.youtube.com/watch?v=0UcTD49KugA

@blackaryz's talk at @normconf about hacking (your own) APIs had a similar takeaway message as my talk about code archaeology: read the manpages of your tools and always try the simplest solution first, and you will amaze your colleagues as you magically solve all their problems.

Next up @normconf: a wonderful talk by generative artist Helena Sarin, who talks about her work with GANs and capturing "the fleeting beauty of early iterations".

Replying to @jd7h

@normconf And this gem of a quote: "A GPU is a girl's best friend."

Replying to @jd7h

@normconf "Early GANs have a sense of humor, they are goofy." ❤️

Preach! The second talk at normconf was @fishnets88 on using common sense over fancy ML tooling. "But you can't buy a online course certificate in common sense and critical thinking." Final message: "We need more personal blogs" with data science and ML stories.

@vboykis opens @NormConf with a machine learning fairytale starring Vectorella. "So Airflow is basically just fancy crontab!" cracked me up.

HUMAN_FALLBACK https://www.theguardian.com/technology/2022/dec/13/becoming-a-chatbot-my-life-as-a-real-estate-ais-human-backup

It's adorable to see ChatGPT try to play Zork. "Senior citizen" Zork fails to understand ChatGPT's "teenage slang". Zork's natural language understanding (NLU) system consists of hardcoded pattern matching, so it can't parse ChatGPT's detailed, fluent sentences.

Replying to @jd7h

"Enter the house" took ChatGPT five tries, as it tried to overspecify how it wanted to enter the house ("through the OPEN WINDOW DAMNIT"). After some nudging/cheating from my side ("try a shorter command") ChatGPT finally managed to get inside...

Replying to @jd7h

ChatGPT found a glass bottle of water inside the white house. Now it's trying to drink the water, but of course it needs to OPEN the bottle before it can DRINK and we're back at square one and it's killing me. 😂

Replying to @jd7h

More info about Zork's natural language understanding can be found here: https://medium.com/swlh/zork-the-great-inner-workings-b68012952bdc and in a repository with Zork I source code (!) here: https://github.com/historicalsource/zork1

Replying to @bert_hu_bert

@bert_hu_bert @ggerganov Lots, probably. I read the code quality was deemed really high by HN readers.

Georgi Gerganov has created a lightweight open source port of OpenAI's Whisper model that can run locally on your CPU. It's fast, fast enough to do real-time transcription of audio. I've tried it out on an old conference talk video from FDG 2021 and the results are impressive!

Replying to @jd7h

Link to the repo: https://github.com/ggerganov/whisper.cpp
Thank you @bert_hu_bert for pointing it out!

I loved reading this interview with @thesephist, who is experimenting with alternative interfaces for AI tools. https://every.to/superorganizers/linus-lee-is-living-with-ai 🧵

Replying to @jd7h

@thesephist 1/ Linus Lee is an independent researcher who want to build better interfaces for interacting with generative AI. He uses AI tools and custom-built models to improve his research productivity.

Replying to @jd7h

@thesephist 2/ The article describes the tools he uses to "live with" AI. What fascinates me is that he essentially created a positive feedback loop between his research subject and his research productivity.

Replying to @jd7h

@thesephist 3/ Lee uses literature review tool Elicit to help him find relevant papers for his research questions. Given a question, it outputs a list of relevant papers and a summary of their abstracts. I'm curious how this compares to GoogleScholar, Arxiv and Twitter recommendations.

Replying to @jd7h

4/ Lee also uses ExplainPaper to help him understand dense papers in his field. If you highlight an equation in a paper, the tool will try to summarize and explain it. Their website doesn't go into detail, but it seems to be based on the Transformer architecture.

Replying to @jd7h

5/ Lee created a tool called CoStructure to help him quickly skim through articles. It creates a visual heat map on top of an article to show the most important sentences using extractive summarization techniques.

Replying to @jd7h

6/ Lee also created a note-taking app called Notation which automatically highlights words in notes and turns them into links. Clicking on a highlighted word yields all other notes with high semantic similarity, sorted by relevance.

Replying to @jd7h

7/ He built Notation with the idea that "no one should have to double-bracket notes", such as in note-taking apps with bi-directional links Roam, Obsidian and Stroll. I like this idea!

Replying to @jd7h

I'm a big fan of knowledge bases with bi-directional links, but consistently bracketing topics is difficult and slows down my note-taking.

Replying to @jd7h

9/ In addition to all of the above, Lee built not one but TWO personal search engines to help him find stuff in his personal data. Monocle is a full text search engine, and Revery uses semantic search.

Replying to @jd7h

10/ The blogpost also features examples of how Lee uses the tools in practice, and how he explores alternative (i.e. non-textual) interfaces in his own tools.

Replying to @jd7h

11/ The interview ends with Lee's main inspiration: two science fiction books! Diaspora by Greg Egan, and Exhalation by Ted Chiang. I'm putting these on my to-read pile because I'm fascinated with the original approach that Lee takes to AI interfaces.

Replying to @jd7h

If you enjoyed this thread:

1. Follow me @jd7h for more of these
2. RT the tweet below to share this thread with your audience.

https://twitter.com/jd7h/status/1601152574186344448

It's been only a few days since the research preview of ChatGPT was released, and it feels like I'm collaborating with an over-enthusiastic research intern.

Replying to @jd7h

I've used ChatGPT as a virtual writing assistant to help me turn my technical notes into full prose text. Yes, it sometimes hallucinates non-factual statements, yes, you have to triple-check each sentence, but editing its output mostly reminds me of grading student papers at uni.

Replying to @jd7h

I've also done some specific & structured experiments to gauge its strong and weak points. Co-created blogpost forthcoming. ;)

The Dutch AI coalition (NL AIC) has created an online course about AI applications in the creative sector. It's fully online, free, and in Dutch. This is a great initiative and it looks really accessible! Register here: https://creatief.ai-cursus.nl/over-ons

Replying to @martinpi

@martinpi @writewithlaika Congratulations! 🚀

Vincent D. Warmerdam @fishnets88

To folks attending @PydataEindhoven; I'm bringing stickers, including some of the rare collectibles from @spacy_io and @explosion_ai!

Thanks @fishnets88 for an excellent talk about hacking unlabelled data into something useful & colorful @explosion_ai swag. https://twitter.com/fishnets88/status/1598307710407217154

Text generation systems "explaining" jokes in a literal, pedantic way is just a new genre in computational humor, right? https://twitter.com/mister_borogove/status/1598110380051816448

Replying to @fishnets88

@fishnets88 @Kay_Hoogland Kom vooral hoi zeggen! Ik hoorde dat je bij Explosion werkt, dat is een van mijn favoriete bedrijven. :D

In preparation for my talk tomorrow, I asked @midjourney_ai what a 'code archaeologist' looks like.

Replying to @mtrc

@mtrc Thanks Mike, please keep the reviews of less well-known games coming. :)

Replying to @martinpi

@martinpi @writewithlaika Ah, the language model has become self-aware! If you need a robopsychologist, I'm your gal. 😁

I'm not procrastinating from writing my PyData Eindhoven talk with datamining projects, what are you talking about.

Replying to @btwsl

@btwsl Yep, I bought them with my first "real" paycheck in 2016.

I'm writing slides for next week's PyData Eindhoven conf, and I've discovered I could probably give a 30 minute talk about git blame alone.

Remember kids, repairing old gadgets is way cooler than buying new gadgets.

Nieuw recept op Mosterdgeel: vegan bloemkoolstoof met chocolade, port en gemberkoek! En ja, dat is hartiger dan het klinkt. https://www.mosterdgeel.nl/bloemkoolstoof-met-chocolade/

Nieuw recept op Mosterdgeel: betere tomaten-crèmesoep. https://www.mosterdgeel.nl/betere-tomaten-cremesoep/

My talk "Practical code archaeology" was accepted for @PydataEindhoven! If you're there, come say hi! :) https://eindhoven2022.pydata.org/cfp/talk/NJVGKJ/

Video game Heaven's Vault nerdsniped me, but now imagine what will happen when I'm able to decode the game's save files and read the entire game state and dictionary in plain text... (No spoilers in the picture)

New blogpost about stockpicking: https://judithvanstegeren.com/blog/2022/investing-in-mergers-and-acquisitions-with-risk-arbitrage.html

"Amateur stockpicking is a dying art, like pie-baking, which is losing out to the packaged goods." --Beating the Street, Peter Lynch

Replying to @mtrc

@mtrc Yeah, I understand. I have my own tight-knit irc/Discord community. But what I like about all those federated instances is that apparently you can follow users from instance X, even if you are on instance Y yourself.

Replying to @mtrc

@mtrc Cohost definitely looks friendly and straight-forward. :)

Replying to @raahatkaduji

@raahatkaduji @thewashistation This looks like something @Spiderwebz might enjoy.

Replying to @mtrc

@mtrc Have you read this primer by Darius Kazemi? It's really good, you might like it! https://runyourown.social/

Replying to @mtrc

@mtrc Why did you pick cohost? I was exploring federated alternatives to social media yesterday, but cohost doesn't seem to be built on ActivityPub?

Judith van Stegeren @jd7h

Amazon says I am not allowed to read over 9000 books in 2020. :(

The tweet below is almost 3 years old. Today I've installed a local development instance of federated GoodReads alternative @BookWyrmSocial to check out the project, and I'm not disappointed.

https://twitter.com/jd7h/status/1219757900962451460

Replying to @jan_czechowski

@jan_czechowski That's a matter of personal preference. Personally I love solving "AI problems" without any AI. ;)

Email from a random recruiter about an NLP job. "Desired experience: 8 years with HuggingFace Transformers"
Me: Yeah, about that...

Replying to @jd7h

"Must have full right to work in UK or EU without sponsorship, and a time machine."

Replying to @lauretig

@lauretig @vboykis I really liked this title, so I'm going to turn this into a talk proposal...

Someone joked that they would give a talk at @normconf titled 'practical code archaeology' but after today I'm seriously considering writing a talk with this title, and extending the archaeology tips to business processes as well.

Replying to @jd7h

Found the original tweet: https://twitter.com/lauretig/status/1552084810751004673

I’m volunteering at Balfolk festival CaDansa, and the volunteers lounge has the best decoration: a garland of chocolate letters

This website is worth reading just for the concise and lucid description of legal issues around training a language model (or other AI systems) on public data. https://githubcopilotinvestigation.com/

NOS Teletekst @Teletekst

Digitale witwasaanpak Bunq mag toch

Dutch online bank Bunq is allowed to implement customer due diligence and anti-money laundering measures with artificial intelligence and data analysis, as ruled by the Dutch Trade and Industry Appeals Tribunal. https://twitter.com/Teletekst/status/1582306557085511685

Momo @momodraws

Please say hi to Snek

Strangely applicable to fintech startups too https://twitter.com/momodraws/status/1574402897445793792

I'm genuinely excited for @BSG_DREDGE, an indie game by @BlackSaltGames that is all your favorite fishing mini-games combined but in an Eldritch horror world. You can play a demo version on Steam already -- just always make sure you're home before dark...

Corollary 1.1: Strive to tweet things that could not have come from GPT-3. https://twitter.com/_RobToews/status/1581735583000694784

@Adarshp fun to hear about your experiences with cycling (infra) in Finland vs California at #euruko. You might like this thread about cycling in The Netherlands too: https://twitter.com/WarrenJWells/status/1573528085051232256

Replying to @DandQ

@DandQ @tomgauld A margin of printers, a serif of typographers, a thesaurus of translators.

Speaker @MelissaKaulfuss has the best slides: everything is an old Mac OS screenshot, slides are notepad and image preview windows. #euruko

Replying to @sergyenko

@sergyenko I will! I’d love to see the streaming setup. 🎛

Really cool talk at #euruko by @thijsc, who is explaining audio engineering (amplification, mixing, compression) with Ruby code snippets.

“We can just ignore those noises from the internet, as long as we create value” —Yukihiro Matsumoto during the #euruko keynote

Replying to @ojmason

@ojmason I read on WikiTravel that the city was rebuilt in the 19th century to look like St Peterburg!

I'll be attending @Euruko with the rest of the Floryn development team, which means I'll be in Helsinki from Wednesday to Saturday. Any Twitter folks in the area? Want to meet up, or share some traveltips for Helsinki? Let me know.

Replying to @SketchesbyBoze

@SketchesbyBoze Let me know if you’re ever visiting the Netherlands.

I tried to find more finance x ML folks on Twitter by looking at @normconf's Twitter followers, and putting all accounts with finance-related words in their bio in one list. It did not result in the high-quality timeline I was looking for, so... #normconf folks, any tips?

Emoji Mashup Bot 🫡 @EmojiMashupBot

🧐 monocle + 😍 hearts-eyes =

This is the perfect "nerd-sniped" emoji. https://twitter.com/EmojiMashupBot/status/1538325712389689345

I'm reading Aggressively Happy by @joynessthebrave, whose tweets made my PhD progress much, much more enjoyable.

Replying to @ojmason

@ojmason Well, I'm secretly talking about using creative AI (which is what we see mostly on Twitter). You can use creative AI just for fun.

This still applies. https://twitter.com/austinkleon/status/1304074600427462659

These "I fed X to an AI and here is the result" tweets drive me crazy. Tell me the size of your dataset, your pre-processing method, the architecture or base model you used! Where did you run it? Is the model black box or not? Was the output what you expected? I NEED TO KNOW

I’m at the PhD defense of Paris Mavromoustakos-Blom for the last defense of the DATA2GAME research project. The thesis is about detecting video game player mental states and emotions, and game personalisation. Shoutout to @ChristophSalge who is part of the committee.

An interesting example of false positives in newspaper article classification, live on the @nrc website. This is an article from 1985 about then-Prince Willem-Alexander attending Prinsjesdag for the first time. Tags include 'film festival', 'schools' and 'marriage'. #NLProc

Yay, employer Floryn now has a fridge filled with Club Mate and Flora Power mate! 🚀

Replying to @m_ou_se

@m_ou_se @Wassasin Oh dear, I just found out that Jellycat has even more crustaceans with party hats. https://www.jellycat.com/eu/search/?term=crustacean

@m_ou_se @Wassasin I saw this, and I thought of you 🦀: https://www.jellycat.com/eu/celebration-crustacean-crab-cc3cr/ #rustlang

Replying to @vboykis

@vboykis 1. From soft to loud/active: Zero7, Psapp, Bonobo, Ratatat, Shpongle, Infected Mushroom, Monolord
2. Or other hummable stuff, such as Iron & Wine.

Replying to @SamMGreer

@SamMGreer @seamoosi I'll check these out! Con artist movies are one of my favorite genres. I also liked American Hustle (and the soundtrack... <3)

Replying to @SamMGreer

@SamMGreer @seamoosi What are the other two?

Interesting thread about power dynamics in pair programming, via https://martinfowler.com/articles/on-pair-programming.html https://twitter.com/sarahmei/status/991001357455835136

Always label your axes.
[Graph title: What percentage of Dutch people eats meat?] https://twitter.com/ionicasmeets/status/1566509318203113474

Not a joke, this is really happening. https://twitter.com/vboykis/status/1552066833582276610

To celebrate that it's September, here a 16th century recipe for butter beer from a Heston Blumenthal cookbook. https://www.theguardian.com/lifeandstyle/2008/dec/14/heston-blumenthal-christmas-recipes

After rave reviews from Rubyist colleagues, I've started reading @sandimetz's "99 Bottles of OOP", also known as "Show me your implementation of '99 bottles of beer on the wall' and I will tell you who you are".

Today I'm learning more about pair programming. Shoutout to @stijnmeurkens for pointing out this excellent guide by @tuple: https://tuple.app/pair-programming-guide/

BabelColour @StuartHumphryes

Travel back 115 years to the Easter of 1917. I have enhanced this autochrome of ladies advertising Kodak film. It was taken during the Great War, at a time when neither had the right to vote, right on the cusp of enormous social change. It is original colour (not colourised).

Hold on to your fabulous hats, ladies of 1917, The Vote is coming... https://twitter.com/StuartHumphryes/status/1564281952072605701

Replying to @iconawrites

@iconawrites I've heard other people recommending Hex by Thomas Oldeheuvelt, about an entity roaming in Beek, a village near Nijmegen in The Netherlands (where I live). I haven't dared to read it myself yet... ;) Apparently the book also won a Hugo award!

Replying to @jd7h

@iconawrites And for something a bit less bloodcurdling I can wholeheartedly recommend The Graveyard Book and Neverwhere by @neilhimself.

As a machine learning engineer I mostly work in Python, but sometimes I have to make changes to our company's Ruby codebases. It's quite a struggle to switch quickly between the two languages (of which one is supernew to me), but this helped: https://www.ruby-lang.org/en/documentation/ruby-from-other-languages/to-ruby-from-python/

Today I’m in the beautiful chapel turned lecture hall at @jadatascience. We’ll challenge the new master students to come up with data-driven business ideas for fintech scale-up Floryn.

Replying to @jan_czechowski

@jan_czechowski I bought these in the first year of my PhD (2018) so they've been with me for a long time. They're very comfortable, but the holes in the toe have become a problem for 6A+ routes. ;)

Out with the old, in with the new! #gripnijmegen #bouldering

Threat Update @threat_update

Fintech mood. https://twitter.com/threat_update/status/1559254537776406533

"2 installed gems you directly depend on are looking for funding. Run `bundle fund` for details."
Can we get something similar, but then for academia? I'm thinking along the lines of "2 authors you cite are looking for funding. Run `biblatex fund` for details."

Replying to @mtrc

@mtrc Congratulations Mike! Great to hear that you got a permanent position!

"We all know that these ESPs harvest our data, but surely they don’t index private email content… right?" https://medium.com/@ryanbadger/magic-links-can-end-up-in-bing-search-results-rendering-them-useless-37def0fae994

My highly esteemed colleague, you must be mistaken. I finished my thesis last year. All necessary research has been conducted, data points have been analyzed, permission from the ethical committee has been obtained. I'm as ready for the battlegrounds as I will ever be.

I'm reading "Penguin by design" by Phil Baines, and I am in love with this vintage Penguin Books poster (1947) by William Grimmond.

"Given choice between complexity or one on one against t-rex, grug take t-rex: at least grug see t-rex"
https://grugbrain.dev/

mike cook @mtrc

Replying to @mtrc

Some of the assorted topics that are on people’s minds at the seminar today. We always add to these throughout the week, although we’ve never had so many on a Monday before!

31 topics for bachelor theses, master theses, PhD theses, grant proposals, and side projects, if you're interested in researching video games, HCI and AI. https://twitter.com/mtrc/status/1538825066984706049

GoogleScholar is taking this opportunity to show off its best stone coal Dutch... "Your essay has received the tenth citation"

Replying to @drtowerstein

@drtowerstein I know the language model is not sentient, but it is good enough to make people BELIEVE it is sentient. Note that even ELIZA in the 1960s was good enough to make people believe this, and you can build ELIZA using only string pattern matching...

I think now is the time to change my job title to 'Robopsychologist'? https://archive.ph/EwHZx

Computer science and linguistics student @AlejandroCiuba has expanded on my video games corpora, and collected additional dialogue data from Hollow Knight! His project, including datasets, can be found online: https://github.com/Data-Science-for-Linguists-2022/Pragmatics-In-Video-Games

@mtrc have you played In Other Waters yet? I think you might like it, and some parts reminded me of your information gathering game.

I've found yet another book recommendation website: https://www.booksread.by/all/janelle-shane
Turns out @JanelleCShane has great taste in NLP experiments AND books.

"Twitch Plays Wizards Chess derailed fairly quickly when the two kings on the board came together and declared a truce because neither side felt like listening to a chat room for maneuvering commands"
Harry Potter x IT department fanfic The Setup Wizard: https://thesetupwizard.tumblr.com

I did not have any specific plans for tonight, but then someone introduced me to a new creative corner of the internet: the Tilde-verse. Origin story here: https://medium.com/message/tilde-club-i-had-a-couple-drinks-and-woke-up-with-1-000-nerds-a8904f0a2ebf

Replying to @jd7h

Where "new" means: new to me, and relatively new compared to other website platforms where usernames (used to) begin with a tilde...

TIL there are literary clocks that tell the time via book quotes. https://literature-clock.jenevoldsen.com/
via https://www.hodinkee.com/articles/this-brainy-online-literary-clock-is-driven-by-jack-kerouac-virginia-woolf-and-haruki-murakami

Today is the last day of #LudoNarraCon. Indie game "The Vale: Shadow of the Crown" really stood out. I have no screenshot to show you because it is a 3D audio action adventure game with a blind (!) main character! It was unlike anything I've ever played. https://store.steampowered.com/app/989790/The_Vale_Shadow_of_the_Crown/

Replying to @drtowerstein

@drtowerstein Beep boop , this thesis was not written by a bot .

Yesterday James Hoffman posted a new video about Kopi Luwak coffee, with this gem inside it.

I'm not working with language data at the moment but I think this discussion is applicable to machine learning on other data as well. Recommended read, if only to keep a nuanced discussion going. https://twitter.com/myrthereuver/status/1516353204958736386

Replying to @myrthereuver

@myrthereuver @annargrs Thanks for sharing this, I also found it a really interesting paper!

Tech company Underlined has a breakout session about using NLP techniques to measure customer happiness and improve customer retention. They extract topics and feedback from client’s customer conversations. #ladiesofdata

Replying to @jd7h

“For topic modelling on conversations, nouns and verbs are most important, for collecting customer feedback, adverbs (and adjectives?).”

Replying to @jd7h

Underlined has moved from being a data analytics company to a product-based company. Because custom-built NLP tools would be too expensive for most companies, they use a standardized setup with a generic layer, a sector-specific layer and a company-specific layer.

Margot Rozendaal from @DPGMediaAI describes the three ways in which this publisher uses NLP: for marketing (personalization, conversion), supporting the newsroom with article categorization based on reader goals, and contextual advertising.

Replying to @jd7h

To determine which ad to show next to what content, advertising is transitioning from static wordlists to a brand-specific approach. Certain topics might be unsuitable context for brand A but acceptable for brand B.

Replying to @jd7h

The word “COVID” was deemed unsuitable for advertising, but then the static, brand-agnostic approach led to problems for publishers because during the pandemic every piece of content was deemed unsuitable for advertising. #ladiesofdata

Data science conf #ladiesofdata is in the beautiful chapel of the @jadatascience building. Participants are checking in while the speaker of @DPGMediaAI is setting up for the first keynote.

Replying to @jan_czechowski

@jan_czechowski I think it's a nice reminder that the technology was around before Bitcoin (2008), and that the distributedness and costly consensus mechanisms are be optional.

Replying to @wcrawly

@wcrawly @HMIutwente And you congrats with starting your PhD! It looks like you have found a topic in which you can pour your creativity. :D

Hot take of the day: Git is a block chain.

What should be the Boolean value of NaN?

I will be attending the #ladiesofdata event at @jadatascience in Den Bosch next Thursday. I'm in luck: this year's theme is NLP and the location is right next door from the office! More info here: https://www.jads.nl/events/ladies-of-data-2022/

"The Stroopwafel is not a food. It is a biochemical dependence addiction device you’ve created and disguised as a modest treat," dixit @joshuakennon. https://www.joshuakennon.com/stroopwafels-have-become-my-obsession/

Replying to @Bram_Ridder

@Bram_Ridder @Floryn_com Yes, you are right. Some ingredients that helped me during job hunting:
- I reserved 6 months to find a job
- I have a huge (social) network, I enjoy talking to peopl, I'm not shy
- I have many friends that work in industry and have the same background
1/

Replying to @jd7h

@Bram_Ridder @Floryn_com - I read "The professor is in" which includes a chapter about leaving academia and translating your skills from academia to industry
- computer scientists are generally in demand
- I have (time to maintain) a blog, Github projects and an active Twitter account
2/

Replying to @jd7h

@Bram_Ridder @Floryn_com And yes I'm aware all of these are the result of major privileges.
3/3.

Replying to @Bram_Ridder

@Bram_Ridder @Floryn_com I don't think it's that special tbh. Most people with a PhD do not stay in academia forever, because there are not enough academic jobs. Just match your PhD skills (research & development skills, communication skills) to industry needs, and voila!

Brought to you by the Center for Humane Technology, the same organisation that co-created documentary The Social Dilemma (viewable on Netflix). https://twitter.com/HumaneTech_/status/1501950899115024387

Happy to announce that since March 1st I'm working as a machine learning engineer at fintech startup @Floryn_com! I love that I can now apply the skills from my PhD to real-world problems. The work is really varied so far, and I'm learning lots of new things from my colleagues.

Nima 🌎 @NimaRoohiS

Hidreley Diao uses #AI to capture what historical figures would look like if they were modern people. 🤩

George Washington:

Note that in this case "uses AI" means "using mobile apps FaceApp, Gradient and Remini to sculpt faces of existing people so they look like statues and paintings"! https://twitter.com/NimaRoohiS/status/1495487546155057153

"Sweet breakfast gravy!" (I'm adding this one to my list of interjections.) All aboard the Ruby train! Excerpt from Ruby Wizardry by @ericqweinstein

Alright alright alright I'll take a look at Eloquent Ruby. https://www.reddit.com/r/ruby/comments/26cxpd/what_is_the_best_way_to_learn_ruby/chq3sgh/

Replying to @jd7h

Eloquent Ruby by @russolsen certainly starts with an eloquent acknowledgments section: "Sometimes I love to write and other times it’s like squeezing out that last bit of tooth-paste—from the point of view of the tube." Very relatable!

Replying to @jd7h

@russolsen This book is hilarious. About formatting conventions in Ruby, and the choice for two spaces: "The trouble with tabs is that the exchange rate between tabs and spaces is
about as stable as the price of pork belly futures."

If you've read all interesting-looking contemporary personal finance books, you can always read historical ones! Wandering through Project Gutenberg books you can find gems like this...

Replying to @jd7h

"People of moderate fortunes have likewise an unquestioned right to dispose of their hundreds as they please; but I would ask, Is it wise to risk your happiness in a foolish attempt to keep up with the opulent?" (1832) https://gutenberg.org/files/13493/13493-h/13493-h.htm#page89

Wow, this is a great introduction to recent developments in Transformer-based NLP, written for a general audience: https://www.forbes.com/sites/robtoews/2022/02/13/language-is-the-next-great-frontier-in-ai/

"Some summarization benchmarks required as few as 1,000 fine-tuning examples for Pegasus to match the performance of Transformer baselines that saw 10,000+ supervised examples — suggesting that one could focus on quality rather than quantity." https://ai.googleblog.com/2022/03/auto-generated-summaries-in-google-docs.html

Replying to @jd7h

This is good news for low-data domains, although in most cases, 1000 high-quality examples is still a lot.

Interested in NLP and legaltech? There a vacancy for a postdoc researcher in the domain of natural legal language processing at my Alma Mater @Radboud_Uni with @mireillemoret. https://www.academictransfer.com/en/309231/postdoctoral-researcher-for-natural-legal-language-processing/

Replying to @mtrc

@mtrc Thank you! Everyone got the impressions that the committee was also enjoying itself. ;)
BTW the ceremony is downloadable/viewable here: https://vimeo.com/event/107744/videos/692285502/
Your webcam feed has been edited into the video in the part of the ceremony where you ask questions!

I successfully defended my PhD dissertation! This was the grand finale of four years of research at @HMIutwente. You may now call me "doctor Judith". 🎉

Replying to @jd7h

The defense itself was a terrific experience, thanks to my amazing support crew @mrngm, @jtvwts and @lorgatti, and the highly esteemed and learned members of my committee @mtrc, @dkjheylen, Franciska de Jong, @EhudReiter, Pieter Spronck, Mariët Theune and @gebwes.

Replying to @jd7h

I'm also thankful to photographer Harm Berntsen for his incredible ninja skills, who made the above photos of the defense ceremony.

Replying to @erwinkooi

@erwinkooi Hahaha, nope, I only have 12 minutes for the layman's talk and after that's it's just academic questions by the committee members.

My PhD defense at the University of Twente is a public event. It will be streamed live today starting at 16:30 CET. Feel free to join if you are interested in text generation or video games research.

* Live stream: https://vimeo.com/event/107744/
* Full thesis: https://jd7h.com/assets/jvanstegeren-flavor%20text%20generation%20for%20roleplaying%20video%20games.pdf

Interesting follow-up to "Google search is dying" with a design for a new ranking algorithm for web pages. Would love to see this in practice, but I don't think using upvotes from Twitter, Reddit and HackerNews will be enough to cover the non-technie web. https://dkb.io/post/market-rank

"When an audience hears acoustic sound, there’s a subconscious box that gets checked that says, this is real." https://www.nytimes.com/2022/03/16/movies/dune-denis-villeneuve-sound.html

Trying to find my bearings in the fintech world now that I'm working at a fintech startup! This primer by @DianacBiggs was useful: https://medium.com/@dianacbiggs/fintech-a-primer-4733405fdd5f

My PhD thesis has arrived from the printer! 😍 Thanks to @jtvwts for collecting the theses at @utwenteEN, and doing a photoshoot while I was at my new job.

My goal for the introductory talk of my PhD defense is 12 minutes of @tomgauld comics.

Look at this cute children's book about the video games industry: https://www.simonandschuster.com/books/If-You-Love-Video-Games-You-Could-Be/Thea-Feldman/If-You-Love/9781534443983

Apparently NBD Biblion, a Dutch organisation that reviews books for libraries and schools, mass-fired their reviewers because they are to be substituted for a natural language processing system. Critical opinion article by one of the human reviewers here: https://www.tzum.info/2022/03/opinie-nbd-biblion-recensent-anna-a-ros-ziedend-over-vervanging-door-kunstmatige-intelligentie-en-roept-op-tot-actie-van-auteurs-recensenten-en-uitgevers/

An interview with the person behind investing education account @10kdiver: https://www.acquired.fm/episodes/10-k-diver-building-a-top-fintwit-account-and-teaching-finance-principles-to-everyone-pseudonymously

@ckaiwu @paradoxinvestor I'd like to build a Twitter list with people that tweet interesting ideas, news, or links about the financial sector x AI, without too much marketing noise. Any tips for accounts to follow?

My PhD defense has been officially announced at the university website -- including a link to the livestream. https://www.utwente.nl/en/education/tgs/currentcandidates/phd/calendar/2022/3/473634/phd-defence-judith-van-stegeren-flavor-text-generation-for-role-playing-video-games

Replying to @jd7h

The definitive version of my PhD thesis is now digitally available: https://jd7h.com/assets/jvanstegeren-flavor%20text%20generation%20for%20roleplaying%20video%20games.pdf

Replying to @APWOTmag

@APWOTmag Aww, that's too bad. I'm looking to extend my Joe Sparrow tarot deck, but I did not back the KickStarter campaign.

@APWOTmag If I order issue 2 via the website, does it come with separate tarot cards? Or is that only for the original kickstarter backers?

Replying to @drtowerstein

@drtowerstein I can send a hardcopy over in exchange for postage :)

So apparently the technical term is "oodles of livestreams"...

Replying to @mtrc

@mtrc @APWOTmag It looks amazing! Halfway through the kickstart page I realised it includes tarot cards by Joe Sparrow and oh dear where is my credit card

Making color variations of my thesis cover design in Adobe Illustrator like a frenzied Andy Warhol. I am my own client from hell. ;)

Replying to @pyrofoux

@pyrofoux Yes! It is a public defense and there will be a live-stream (starting at 16:30 CET) for people that would like to watch remotely.

I'm happy to announce that my dissertation "Flavor text generation for role-playing video games" was accepted by the thesis committee. Onwards! to the defense ceremony on March 25th.

Now this is a description of one person's investment practice that I like to see: http://roambrain.com/roam-investing/ The blogpost focuses on note-taking app Roam. I'm currently trying out a similar app (but opensource, free and local) called Stroll.

Interesting to see a technical explanation of the back-end of all these hip aleatoric portraits. https://twitter.com/nyx_eth/status/1472381291995799552

Thesis formatting horrors no one will tell you about:
- the 579898 alternatives to bibtex
- having ONE bib-entry with non-English-language characters in the author name
- The Exact Ordained Order of importing LaTeX packages because otherwise undefined or unexpected behaviour.

Replying to @jd7h

And don't forget the intricate & undocumented LaTeX styles for dissertations, passed down as departmental heirlooms through generations of PhDs.

I spend a lot of time thinking about how I can efficiently organise and search all the information in my head, let alone the information on the internet. Interesting and relevant read: https://dkb.io/post/organize-the-world-information

I received an email today from a student that wants to work with video game assets, and maybe create a new video game (text) corpus. They had some questions about copyright and Fair Use, and I referred them to this article from the DiGRA library: http://www.digra.org/wp-content/uploads/digital-library/ScreenshotsFairUseRecommendations_DiGRA.pdf

A high-quality 2-hour explainer video (documentary? investigative journalism? distopian non-fiction? rant?) about N\FTs and cry\ptocurrencies: https://www.youtube.com/watch?v=YQ_xWvX1n9g

A Short Hike is a casual indie game I would recommend to anyone. It has nature, hiking, flight mechanics, friendly NPCs, hilarious dialogue, bright colours, voxel art and... an outdoor rock climbing wall! 10/10

Replying to @ojmason

@ojmason Haha, als je een goed recept met aardvark weet, hoor ik het graag! ;)

Accidentally discovered this morning that the first recipe of my online cookbook http://www.mosterdgeel.nl (in Dutch) was published a bit more than 10 years ago. 🎉

This was completely unplanned, but I got interested in the superwacky names of famous race horses (long story) and now I've spent the day learning my way around Wikidata and SPARQL. I now know how to write queries to get massive linked datasets from Wikipedia... ✨

Replying to @jennysaucerman

@jennysaucerman After your tweet, I thought up the combination of TEAMS, HOUND, QUERY and LIGHT.

Playing Wordle is easy once you've used English letter frequency lists to manually decode substitution ciphers for your first-year course in computer security... or you could read this page: https://norvig.com/mayzner.html

I'm halfway through my two-month mini-sabbatical, and I love that I can binge-read every single day, with no interruptions and lots of espresso. Apart from a final read-through of my thesis manuscript, I don't have anything else to do. This is the life! 😎