tokenization on S Anand

Things I Learned - 31 May 2026

Sun, 31 May 2026 00:00:00 +0000

This week, I learned:

D-ID is an avatar generator platform like HeyGen. Creatify and Synthesia are a couple of others I heard of. This space seems to be growing.
cosign is a CLI that lets you sign and verify any piece of text with a Google, GitHub or Microsoft account. cosign sign-blob FILE --bundle sign.json opens a login window and creates a sign.json signature. Anyone who has FILE and sign.json and the email ID can verify via a Google account with cosign verify-blob FILE --bundle sign.json --certificate-identity $EMAIL --certificate-oidc-issuer https://accounts.google.com.
arxiv2md.org converts arXiv papers to Markdown. Source. markxiv.org claims the same - by just changing the URL - but it ended up reporting an error when I tried this link: https://markxiv.org/abs/2604.08649.
From Akhilesh Tilotia: So we have someone in our team with initials AS. She made a document which was named vAS. Then I made edits and named it vAT. These docs were in a CoWork folder. I asked Claude to clean up my doc. It created another version for me to review. In its wisdom, it named the file vAU 🙂
Maybe what a forward-deployed engineer does is enginer AI-native workflows. (This sounded profound when I wrote it down. Not sure if it’ll sound as profound tomorrow.) The idea is that the FDE will say, screw existing processes; let me fire up my AI agent and get stuff done; THEN we’ll figure out what works, how to optimize it, etc.
The PRAGMA: Revolut Foundation Model has some good tokenization ideas for tabular data. Create your own token space with key–value–time tokenization - to retain field information. Bucketize numbers by percentile, preserving magnitude/ordering that subword tokenization destroys. Encode time both as log-seconds and as cyclical calendar features.
Codex uses the Alt + Up Arrow key to edit queued commands, but on the VS Code terminal, this key binding is not sent to the terminal. Enable the terminal.integrated.sendKeybindingsToShell setting to send it to the terminal, hence Codex.
Based on this catalog on “universal foods”, here’s what I 🟢 like, am 🟡 neutral, 🔴 dislike, 🟣 must try, and will ⚫ skip.
- Universal favorites: 🟢 pizza, 🟢 fried potatoes/chicken, 🟡 dumplings, 🟢 ice cream.
- Universal comfort foods: 🟢 khichdi, 🟡 congee, 🟡 dal-rice, 🟡 risotto, 🟡 ramen, 🟢 pho, ⚫ chicken noodle soup, 🔴 rice porridge, 🟡 mac-and-cheese, 🔴 mashed potato, 🟣 polenta, 🟢 oatmeal, 🟣 Japanese curry rice.
- Acquired tastes that convert most: 🟡 coffee, 🟢 tea, 🟡 dark chocolate, 🟢 mild fermented dairy, 🟢 pickles, 🟢 olives, 🟣 kimchi, 🟣 miso, 🟢 mild chili dishes.
- Acquired tastes that have cult devotion: 🟣 durian, 🟣 natto, 🟣 stinky tofu, ⚫ fermented fish, ⚫ hákarl, 🟢 very funky blue cheese, ⚫ offal.
OceanoPDF seems like a good place to download ePubs of books.
The entire Wikipedia is available as a Parquet file. You can query it like duckdb -c "FROM 'hf://datasets/wikimedia/structured-wikipedia/enwiki/data/*.parquet' LIMIT 5". The English version has 35 GB, 7.6 million articles, and you’re better off downloading it rather than running analyses remotely.
When you receive a Calendly link of the form https://cal.com/USER/EVENT you can fetch the available slots via curl -H 'cal-api-version: 2024-09-04' 'https://api.cal.com/v2/slots?eventTypeSlug=EVENT&username=USER&start=2026-05-25&end=2026-06-01&timeZone=Asia/Singapore&format=range'. Useful to automate good meeting-slot selection.
“Reference saved memories” in ChatGPT is different from “Reference chat history” as per OpenAI. In Developer Mode, memory is turned off, but not chat history. I confirmed that I can access past conversations in Developer Mode. It might be a privacy concern for others, but for me, this is singularly useful, because I can use ChatGPT with Local MCP effectively getting a non-metered AI coding agent.
Seems GPT-5.2 reaches expert level in peer review: 45 scientists took 469 hours evaluating human & AI reviews on 82 papers. “Surprisingly, current AI reviewers are competitive even with the top-rated reviewers in Nature’s official peer review…” though not without weaknesses, so use AI + humans. On the limits and opportunities of AI reviewers: Reviewing the reviews of Nature-family papers with 45 expert scientists via Ethan Mollick

Things I Learned - 25 Feb 2024

Sun, 25 Feb 2024 00:00:00 +0000

This week, I learned:

Architecture.md is an emerging standard
Managing wealth requires training.
htmz is a fantastic way to load HTML into elements!
Suguna Poultry is
- Using robots to walk in their farms, use sound and bird eyes and movement to predict birth health over 1-2 weeks
- Light on the back of the bird’s back AND face => lays eggs in 14 days, else takes days later (girls and mobile phones?)
Teknoturf is using Gen AI to
- Improve prompts when teaching prompt engineering.
- Pronounce languages better, identifying which words Tamilians and Malayalis will mis-pronounce.
Explore IRBlaster. It can control AC and can automatically increase temperature at night.
My view: LLMs are general purpose and more capable than SLMs. They’ll win, like CPUs won over special-purpose chips. GPUs will optimize for LLMs and as usage grows, cost will fall.
Andrej Karpathy’s summary of sharp edges in tokenization uses tiktokenizer to explain:
- Why LLMs can’t be used for spelling
- Why LLMs are better at English than other languages
- Why LLMs are bad at math
- Why SolidGoldMagiKarp is a single token
- Why trailing spaces are bad
- Why YAML tokenizes more efficiently than JSON
ssyoutube.com: Just add “ss” to “youtube.com” on the video and you can download YouTube videos
Discussions with Sachin, AMAT
- Microsoft said Indigo, Air India uses LLM based bookings
- Meta invested $70bn in GPUs. Sam Altman is investing $7tn!
- NVIDIA has a price PREMIUM not discount for bulk GPUs!
- AMD is the next company to watch for
- Numenta - Subutai Ahmad - deploys AI models on CPUs
- #TODO Read A Thousand Brains by Subutai Ahmad
- Sanjeev Sharma
  - Swaayatt Robots: Autonomous driving in India
  - Deepeigen: Education
- Rohan Shravan, Bangalore.
  - Likes sharing knowledge. Amazing teacher. IIT KGP 2008. Interested in exploring quantum computing
  - Tresa Motors, Inkers App, The School of AI
- AMAT is working on
  - photon-based computing.
  - science research models. AI for science. Like Google: Deepmind Genome, Microsoft: Metagen
  - quantum: AMAT is actively in into this. Nagapati Banda is driving this
- John Kelly is predicting a ChatGPT moment in quantum in a few years
Adobe express has a forever free video to GIF converter
Edge workspaces let me keep the same tabs open across laptops!
Command line interface guidelines
RAWGraphs has a custom charts API that is worth learning from
Python fastcore has decorators like @typedispatch, Self, etc.
All image-to-text models on HuggingFace
wddbfs mounts SQLite as a file system. I had a bit of trouble, maybe with Python package versions.
Google is using LLM powered bug identification
HuggingFace Chat Assistants has open source system prompts!!
OpenHermes training dataset is available. 1M prompts!
Jio has made IPL free. They make money on data and ads. That’s Scale!
Daniel Dennett outsources thinking to students. Reviewing his books.
- BUT: I don’t take feedback. When someone sends a pull requests, I ignore it.

Splitting a sentence into words

Thu, 26 Apr 2007 12:00:00 +0000

I often need to extract words out of sentences. It’s one of the things I used to build the Statistically Improbable Phrases for Calvin and Hobbes. But splitting a sentence into words isn’t as easy as you think.

Think about it. What is a word?

Something that has spaces around it? OK, let’s start with the simplest way to get words: split by spaces. Consider this piece:

"I'd look at McDonald's," he said.
"They sell over 3,000,000 burgers a day -- at $1.50 each."
High-fat foods were the rage. For e.g., margins in fries
were over 50%... and (except for R&M & Dyana [sic]) everyone
was at ~30% net margin; growing at 25% too!

Splitting this by spaces (consider new lines, tabs, etc as spaces too.), we get the following:

"I'd
look
at
McDonald's,"
...

Now, some of these like “I’d” are words. But “McDonald’s” isn’t. I mean, there’s a full-stop and a double-quotes at the end. Clearly we need to remove the punctuation as well. But, if we do that, I'd becomes Id. So we need to be careful about which punctuation to remove. Let’s take a closer look.

The following punctuation marks are clear word separators: spaces, the exclamation mark, the question mark, semicolon, brackets of any kind, and double-quotes (not single quotes). No word has these in the middle. If we use these as separators, our list of words is better, but we still have some words with punctuation:

McDonald's,
e.g.,
High-fat
R&M
...

The issue is, these punctuation marks are ambiguous word separators: comma, hyphen, single-quote, ampersand, period and slash. These usually separate words, but there are exceptions:

Comma: Not inside numbers: 3,000,000.
Hyphen: Not for hyphenated words: High-fat.
Single-quote: Not for possessives: McDonald’s. Not for joint words: I’d.
Ampersand: Not for abbreviations: R&M
Period: Not for abbreviations: O.K. Not for URLs: www.s-anand.net
Slash: Not for fractions: 3/4. Not for URLs: google.com/search

Colon is ambiguous too. In normal English usage, it would be a separator. But URLs like http://www.s-anand.net/ use these characters, and it doesn’t make sense to separate them.

So here are my current rules for splitting a sentence into words. (It’s a Perl regular expression. Don’t worry. Cooper’s Law: If you do not understand a particular word in a piece of technical writing, ignore it. The piece will make perfect sense without it.)

# Split by clear word separators
/       [\s \! \? \;\(\)\[\]\{\}\<\> " ]
 
# ... by COMMA, unless it has numbers on both sides: 3,000,000
|       (?<=\D) ,
|       , (?=\D)
 
# ... by FULL-STOP, SINGLE-QUOTE, HYPHEN, AMPERSAND, unless it has a letter on both sides
|       (?<=\W) [\.\-\&]
|       [\.\-\&] (?=\W)
 
# ... by QUOTE, unless it follows a letter (e.g. McDonald's, Holmes')
|       (?<=\W) [']
 
# ... by SLASH, if it has spaces on at least one side. (URLs shouldn't be split)
|       \s \/
|       \/ \s
 
# ... by COLON, unless it's a URL or a time (11:30am for e.g.)
|       \:(?!\/\/|\d)
/x;

This doesn’t even scratch the surface of the issue, though. Here are some issues:

Lots of files split words into two at the end of a line. How do we handle that?
How do we handle incorrect punctuation? For instance, if someone types “done.Yet,” without leaving a space after the full-stop, I’ll think it’s an abbreviation.
What about other separators? Like the ± symbol or the £ symbol for instance.
What about other languages?!

And you thought it was easy!

Comments

Kasi 27 Apr 2007 10:18 pm: Hi Anand, Nice site..It was good talking to you! When u find time, pls do visit my humble less visited blog at www.raconteurkasi.blogspot.com !
Ramya 3 May 2007 7:52 am: Hi Anand, Can u guide me through the interview process of INSEAD MBA . Can you share your wisdom with me to crack the interview, and one more question .. do u sleep for less than 5 hrs a day ?
Andrew 26 Apr 2007 12:00 pm: Thanks, this is very helpful.

Statistically improbable phrases 2

Wed, 14 Feb 2007 12:00:00 +0000

My earlier list of statistically improbable phrases in Calvin and Hobbes is technically just a list of “Statistically Improbable Words”. I re-did the same analysis using phrases. Here are the top 20 statistically improbable phrases (2 - 4 words only):

baby sitter chocolate frosted sugar bombs comic books doing homework fearless spaceman spiff() good night hamster huey ice cream miss wormwood new year peanut butter really think slimy girls spaceman spiff stuffed tiger stupendous man sugar bombs susie derkins watch tv water balloon

That is, these are the 2-4 word phrases whose frequency in Calvin and Hobbes is substantially (at least 5 times) higher than in the other books I have.

While doing this, the single biggest problem that stumped me was: what is a word?

Is “it’s” one word or two words?
Is “six-year-old” one word or three words?
How do I distinguish between abbreviations (g.r.o.s.s.) and full-stops without a space ( … homework.what’s a …)?
Does a comma always split words? (It doesn’t in numbers, like “3,500”)

The other problem is, phrases with more words are more improbable. Right now, if a phrase occurs 5 times more frequently in Calvin and Hobbes than my other books, I include it. But three-letter words rarely occur that often, and four-letter words even less so. Maybe I should have a lower cutoff for longer phrases.

Anyway, this analysis is a crude first approximation. Clearly Amazon’s gotten much further with their system.

Comments

satish 20 Feb 2007 3:56 am: Hey Stud, Satish here, your junior from IIMB. Trying to get in touch with you. Do mail me at satishkgv@hcl.in and let us get in touch.
Oblio 14 Feb 2007 12:00 pm: Fantastic job man! You have unlimited patience!
Reinhard Ebner 14 Feb 2007 12:00 pm: Hey, only just now came across your page, but of the hundreds, if not thousands of C&H sites and tools, this is the most useful I’’ve seen! R
juergwachter 14 Feb 2007 12:00 pm: hello
nice stuff.I intend splitting a text into single words. can you please give me a hint how to do this? I guess there are simple programs doing this.
Many thanks
joe 11 Aug 2009 2:01 am: Nice. Do you have a page where I can try out v2 (phrases)?
The Calvin and Hobbes search Takedown | s-anand.net 21 May 2010 11:53 am (pingback): […] was able to do a lot of cool stuff with this, like statistically improbable phrases and many amusing […]