behavioral-economics on S Anand

Things I Learned - 25 Jan 2026

Sun, 25 Jan 2026 00:00:00 +0000

This week, I learned:

POSSE - “Publish (on your) Own Site, Share Everywhere” - is a self-explanatory content sharing approach. 1 minute video introduction. Alternatives are:
- COPE: Create once, publish everywhere.
- POSE: Publish once, syndicate everywhere.
- PESOS: Publish elsewhere, syndicate to own site.
- PESETAS: Publish elsewhere, syndicate everything to a silo (one of the “elsewhere"s).
Cory Doctorow’s essay on how Google will use AI against shoppers is a great lesson on several lessons in behavioral economics. Fighting it is hard, but here is generally good advice: ChatGPT
- Never buy when rushed.
- Set a max price.
- Compare 2+ options (TOTAL price) early.
- Shop logged-out for big purchases.
Vortex is supported by DuckDB. It’s a better format than parquet for analysis and querying remotely. Gemini
Research suggests that insight emerges when we struggle with a problem, get stuck, then DISENGAGE (to inhibit distractions). EEGs can predict insight ~8 seconds before it happens using this pattern. Gemini
- Books that have suggested this are:
  - The Art of Thought (1926)
  - The Act of Creation (1964)
  - The Inner Game of Tennis (1974)
  - Hare Brain, Toroise Mind (1997)
  - Taoism’s Wu Wei: “muddy water, let stand, becomes clear”
  - Zen Koans: giving students impossible riddles
- Books that have argued (probably incorrectly) that an impasse is a signal to push harder, not step back, are:
We are starting to talk like LLMs. Empirical evidence of Large Language Model’s influence on human spoken communication
Email support for animated AVIF is limited. Though it’s perhaps the best compression format, and Apple Mail supports it well, Outlook on Windows does not support it. GMail converts it to GIF. Also, Google Groups does not yet support it. When I sent an animated AVIF in a Google Group email, it was replaced by the content proxy to this nonexistent URL.
Portkey Models is a repo of model related data (e.g. price, max tokens, capabilities, etc.) for a large number of models. Somewhat similar to Simon Willison’s LLM Prices.
git-filter-repo is a surprisingly easy way to rewrite your git history to remove specific files (e.g. large files, sensitive files). It can preserve timestamps, messages, etc. It just filters out specific files as if they were never committed. This can reduce the size of repos dramatically.

Things I Learned - 24 Aug 2025

Sun, 24 Aug 2025 00:00:00 +0000

This week, I learned:

Pilots like to have fun, too. While awaiting landing clearance at Kolkata, our IndiGo pilot weaved tight curves just above the clouds at steep angles, giving us stunning views and a mildly thrilling experience. (Or maybe they were just following a flight path.)
Since LLMs allow ANYONE to become “good enough” in most fields (marketing, medicine, management), and so on, here’re are my guesses on the impact. ChatGPT
- Companies-of-one will grow. Sole founder can handle support functions.
- Specialists will generalize. Consultants will code. Marketers will design.
- Wages will compress. Seniors will earn less as juniors can do more.
- Layers will compress. Organizations need fewer hierarchies as 1 person can do more.
- Shadow apps will grow. Anyone can code. Users build apps with prompts, sheets, agents, outside of IT SDLC. Like Excel sheets.
- Governance will grow. Non-experts are acting like experts. Validation is more important.
- Uneconomical apps will thrive. 1:1 tutoring. Continous decision making or A/B testing.
- Leaders will convince better. Persuasion scales.
- Brand (authenticity, trust, skill), Channel (distribution, audience) and Data are primary differentiators.
Codex and Codex CLI now support image attachments.
Notes from discussion on education with Srikanth Nadhumuni
- Indian higher education has done better, e.g. with the IITs, than primary education, where ASER consistently shows that 5th graders can’t read 2nd grade books.
- The National Education Policy (NEP) is focusing on FLN (foundational numeracy and literacy). The goal is universal FLN by 2027.
- Teacing FLN in local languages beats English. Teachers, parents, community support are high. Learning English as a second language is faster. Other countries (France, Germany, Japan) do this.
- Voice LLMs could help, but may not be toddler-ready, nor strong enough in all local langauges.
- But high-quality textbook translation with local nuances is a one-time human-in-the-loop effort that AI can support.
- India’s 1 crore teachers have a mandatory 50 hrs/year training requirement that is largely under-implemented.
- Senthil Mullainathan is working on extracting features from student answers to questions and generating remedial content purely as a black-box. Results beat explainability.
⭐ Creating systems that rapidly improve from feedback is the key to success. Rapidity, quality of improvement, quantity of feedback are all enablers.
CBDC (Central Bank Digital Currency) is RBI’s Web 3.0 protocal. It allows purpose-driven transfers, e.g. money meant for education can only be spent on education.
Meta-prompts with placeholders is a prompt-improvement technique (similar to LLM interviewing). Have LLMs create the prompt with “fill-in-the-blanks”. This makes it much easier for people to fill out.
MassGen is a multi-agent orchestrator. Early days, experimental. It has multiple agents answer, then vote on each others’ answers, picking the best.
DSPy auto-optimizes prompts based on input-output pairs or evals. Typical improvements are ~10-20%. My opinion: avoid. It’s a good idea, but has too much abstraction that hides the implementation. Worth learning from but not implementing unless you (a) have evals + metrics and (b) you KNOW you need to change models and (c) it’s a long-term project where the learning curve is worth it. Claude and ChatGPT
How LLM “Attention” works: It takes each word’s embedding, moves it closer to similar words’ embeddings (e.g. Apple moves towards phone or orange depending on context). More similar words have a higher pull, like gravity. Luis Serrano
- Similarity isn’t symmetric. E.g. “Coke” moves “drink” more towards it, but “drink” pulls “Coke” less, since “drink” could refer to other things.
- Think of the pull (“Tinder similarity”) as “what A wants” (key matrix, which pulls other words) multipled by “what B offers” (query matrix, which is pulled by other words). This leads to two different similarity matrices.
- Multi-head attention is where a neural net gives different weightages to different similarity matrices based on context.
- Value matrix transforms the embedding space so that the next best next-word is more similar.
Reading the Obsidian docs is like a master class in Markdown note-taking. Features like properties, embedding YouTube, bases, tags, etc. provide food for thought. The ObsidianMD subreddit has interesting tips.
- Summarize takeaways on top of each section
- Use atomic notes: one file per idea. Link liberally
- YAML front-matter you can query, e.g. tags, project, status, …
- Use GFM admonitions, e.g. > [!NOTE]
- Store images in a predictable way, e.g. ![Alt text](./img/2025-08-21-screenshot.webp) – ALWAYS with alt text
- Use diff fences for edits / doc changes
- Task lists with inline dates, e.g. - [ ] 2025-08-21 Draft a letter
How to research better. Abhishek Divekar
- Have an objective when researching. Filter research based on that.
- Research backwards. Pick a relevant paper. Go through relevant citations. Typically, there are only 1 or 2 directly related ancestors.
- Don’t waste time searching. Gemini Deep Research is a great way to find and read papers.
- Don’t read the abstract. Read the introduction, which is the summary. It’s just a page. (The abstract is an LLM-ized versionof the introduction. Not as effective.)
MCPs aren’t much more useful than tool calling for developers. They’re powerful when packaging for external parties (non-developers, other teams, clients, etc.). Developers can work just fine with tool calling. Nitin Agarwal
Cybersecurity AI is an open-source LLM-based cyber-security tool that auto scans networks for vulnerabilities.
⭐ LLMs have solved several complex tasks (e.g. topic modelling, summarization). We need to adopt these as building blocks, like functions, and build better solutions. Abhishek Divekar
codex -c model_reasoning_effort=high lets you run Codex CLI with highest reasoning effort. This has a separate limit that resets every 5 hours. https://x.com/thsottiaux/status/1958035261947781262
Truly agentic systems have high Autonomy, Complexity, and Reliability. Workflows have low autonomy. Agentic systems with high autonomy currently aren’t very complex or reliable, but will improve over time. Deepak Sharma
Allow humans to intervene while agent loops execute, even unsolicited, to improve collaboration. Deepak Sharma
Given the early, experimental days of AI, the better KPIs might be more about experimentation (e.g. number of prototypes) than operational (e.g. cost reduction). Krishnakumar Menon
⭐ Policy-as-code is an emerging theme. Allow users to create their own guardrails policy. Or, take existing policy documents and convert them into an LLM-based evaluator. Krishnakumar Menon
⭐ “Potentially nitpicky but competitive advantage in AI goes not so much to those with data but those with a data engine: iterated data aquisition, re-training, evaluation, deployment, telemetry. And whoever can spin it fastest. Slide from Tesla to ~illustrate but concept is general.” Andrej Karpathy, Dec 2022
The skills AI coding needs are very similar to tech-lead’s or an architect’s. Tanika Gupta #ai-coding
- Estimating tool capability & task allocation
- Task breakdown
- Spec-ing: which of user personas, user-journey maps, wireframes, technical architecture, psuedo-code
- Standards: tech stack, tools, linters, security, doc standards
- Git versioning & collaboration
- Code review. (Using AI.) Providing feedback. Modularity, naming, …
- Automated validation
- Post-mortem. Learning from errors and successes, choices LLM made
The ROI of prompting carefully and using meta-prompts is high. Prompt clarity reduces iterations & dead-ends. The initial time spent (10-15 min) pays off with just a single reduced iteration (time to generate + review). Tanika Gupta
⭐ Prefer passing a spec.md to AI coding agents rather than directly typing-in prompts. This lets you meta-prompt and (collaboratively) iterate on the spec.md, version the prompts as specs, and generate specs as documentation. Tanika Gupta
⭐ Models need environments to learn. So far, we have been providing training data. But an environment to interact with, and learn from by itself, is more powerful. That requires a standard for environments. This is a powerful emerging area.
The crux of experimentation is the learning from a postmortem. From that perspective I have been experimenting a lot but not been documenting or learning from that. Decision logs with post mortem are a more apt device for me.
Gemini API includes a url_context tool to explicitly scrape websites. API
Ontologies are more than taxonomies or schemas. They’re truths or rules, e.g., “no person has more than two parents”. Helps consistency checking and inference. #
- Terminological knowledge (T-Box) is domain rules and constraints (e.g., “a student is a person who attends a course”).
- Assertional knowledge (A-Box) is instance-level facts (e.g., “Mary attends Physics 101”).
- Tools & Formats
  - SHACL. A W3C language for validating RDF graphs. ShEx is easier ad popular.
  - Notation3. A W3C assertion and logic language which is a superset of RDF.
  - EYE Reasoner. Prolog-based N3 (Notation3) reasoner. CLI + API-friendly. Can perform rule-based reasoning and generate new triples.
  - HermiT. OWL 2 DL reasoner. Can check consistency, classify ontologies, compute entailments. CLI and Java API. Modern, maintained.
  - Apache Jena. Java framework for RDF/SPARQL. Built-in reasoners (RDFS, OWL mini/micro/full). CLI via riot, arq (SPARQL query engine). Popular for RDF graph stores + inference.
Do developers feel this way? #ai-coding
In another example of vibe coding, an instructor for my TDS course vibe-coded most of an exam using Copilot and Sonnet. 6/8 questions worked one-shot. The two #ai-coding failures were interesting:
- One failed because of sample vs population stats. Copilot asked for sample variance but coded variance() instead of sampleVariance().
- Another failed because of rounding off. NumPy code rounds off differently from Python or JS code.
Meditation is about noticing distraction and returning to focus. So, distraction is necessary and good. #beliefs
#ai-coding can make us overconfident. (At least, it makes me overconfident.) They create surprisingly good output, but only ~20% of the time. I cannot commit to a specific task based on that. Instead, it’s better to rely on AI coding estimates for portfolios, e.g. promise to share something cool without mentioning what. Or do something cool first, then share.
Notes from podcast with Daniel Kahnemann. The Knowledge Project.
- Happiness is pleasure in the moment. Satisfaction is the meaningful story of our life. When we think, we want satisfaction. When we feel, we want happiness. The thinking brain and feeling brain optimize for slightly different things. E.g. The thinking brain packs the calendar with satisfying tasks that the feeling brain feels unhappy executing Both are good for us. We don’t know which matters more.
- Behavior change is harder than we think. Usually, it’s better not to expect success in changing others, or ourselves. Instead, understand why that behavior makes sense. Our behaviour is an equilibrium of forces. Weakening “bad” forces is easier than strengthening “good” forces, since it lowers tension. That’s inversion!
- Behaviours tell us more about situations than personality. We assume otherwise. That’s an attribution error.
- Motivation is complex. People can do bad things for good reasons and vice versa.
- “Feelings get in the way of clear thinking.” Example: I vibe-coded the last 2 questions of TDS GA7 on Claude Code. It didn’t run. I delayed fixing it for 5 days, afraid it would a major effort. It ended up a 2 min fix. It could have been major, but checking would have helped. Fear prevented that.
- Things that hamper clear thinking: intuition, emotion, beliefs. Beliefs are often formed based on people we admire or identify, not reason.
  - Prefer rules, systems and processes. Willpower is an illusion.
  - Delegate decisions to unemotional agents. (But agents misjudge perceived value of gain or loss!)
  - Break down the problem, analyze it, THEM form an intuition. Be disciplined in delaying intuition or forming an opinion
- Environment shapes thinking but it’s not obvious how, e.g. some people work better in noisy cafes. Some colors are more calming.
- Protect dissenters and dissent. It’s painful and costly, and needs nurturing.
NodeJS runs TypeScript files natively.
Codex can clone any GitHub repo. So I can ask it to pull one or more repos, understand their code, and use that as a template or reference. This makes my repositories (and others’) reusable templates. Using newer libraries and platforms becomes easier, too. #ai-coding
Tracking AI runs an IQ test on various LLMs every week. GPT 5 Pro leads, currently, followed by Claude 4 Opus and Gemini 2.5 Pro. It’s surprising how far behind GPT 5 is at the moment.
LLMs are faster than me. So me learning and doing what the LLM says is a bottleneck. Get out of the way. For example do not learn. Do not execute. Do not verify. Give LLMs the tools to deploy, verify and iterate to improve.

Resolving the Prisoners Dilemma

Sun, 17 Aug 2008 12:00:00 +0000

If you’re ever taken a course in Economics, and it discussed Game Theory, you may be familiar with The Prisoner’s Dilemma. Roughly, this is the problem.

Assume you possess copious quantities of some item (money, for example), and wish to obtain some amount of another item (perhaps stamps, groceries, diamonds). You arrange a mutually agreeable trade with the only dealer of that item known to you. You are both satisfied with the amounts you will be giving and getting. For some reason, though, your trade must take place in secret. Each of you agrees to leave a bag at a designated place in the forest, and to pick up the other’s bag at the other’s designated place. Suppose it is clear to both of you that the two of you will never meet or have further dealings with each other again.

Clearly, there is something for each of you to fear: namely, that the other one will leave an empty bag. Obviously, if you both leave full bags, you will both be satisfied; but equally obviously, getting something for nothing is even more satisfying. So you are tempted to leave an empty bag. In fact, you can even reason it through quite rigorously this way: “If the dealer brings a full bag, I’ll be better off having left an empty bag, because I’ll have gotten all that I wanted and given away nothing. If the dealer brings an empty bag, I’ll be better off having left an empty bag, because I’ll not have been cheated I’ll have gained nothing but lost nothing either. Thus it seems that no matte what the dealer chooses to do, I’m better off leaving an empty bag. So I’ll lease an empty bag.”

The dealer, meanwhile, being in more or less the same boat (though at the other end of it), thinks analogous thoughts and comes to the parallel conclusion that it is best to leave an empty bag. And so both of you, with your impeccable (or impeccable-seeming logic), leave empty bags, and go away empty-handed. How sad, for if you had both just cooperated, you could have each gained something you wanted to have. Does logic prevent cooperation? This is the issue of the Prisoner’s Dilemma.

There’s nothing wrong in the logic, actually. The key assumption is that it is clear to both of you that the two of you will never meet or have further dealings with each other again. If you’re never going to deal with someone again (and hence there is no question of retribution or any fallout), you really should cheat.

An aside. During my first few days at IIT, two third years were ragging me about my stance on pre-marital affairs. After trying my best at defending the moral standpoint, I finally confessed that it was only the fear of the after-effects that worried me.

“There’s this beautiful naked girl in your room,” they said. “You are guaranteed no repercussions. What will you do?”

“No repercussions?”

“None whatever.”

I thought for a while. “I’ll flip a coin.” :-)

Anyway, the aside aside, the solution to the one-off Prisoner’s Dilemma is for both people not to cooperate. If contracts are not enforceable, we’re all better off not trading. If there are no cops, we’re individually better off stealing.

But, of course, that’s not true in the real world. Most situations are repeatable, and you do tend to meet people again. Those you cheat may even have a motivation to meet you again.

This is the iterated Prisoner’s Dilemma. Douglas Hofstadter wrote about this in the May 1983 issue of Scientific American in his Metamagical Themas column (which, by the way, is brilliant). While the Prisoner’s Dilemma has a simple solution (don’t cooperate), the iterated Prisoner’s Dilemma does not have a predetermined solution. At best, you can have a strategy. (That can be proven. If there was a predetermined solution, your opponent would know it, and could beat you. One of those cases in mathematics where you are not guaranteed a solution.)

But is it possible for cooperation to emerge in an iterated Prisoner’s Dilemma? Can it beat competitiveness?

Robert Axelrod of U.Mich conducted a computer tournament to find out. He invited strategies from game theorists and wrote them as BASIC programs. Each program would be pitted against another. Every time, it could response with either C (cooperate) or D (defect). Cooperation gets both programs 3 points. If one defects, the defector gets 5 points and the cooperator gets nothing. Both defecting gets 1 point each. Axelrod ran each program against each other many times, and added up the scores.

The program that won was called TIT FOR TAT. It was the shortest program (4 lines of BASIC code). Here’s it’s strategy:

Cooperate the first time.

Do what your opponent does thereafter.

Think about it. TIT FOR TAT starts by being nice, and stays that way, unless you defect. Then TIT FOR TAT punishes. If you repent, TIT FOR TAT forgets and forgives. Interestingly, TIT FOR TAT can never win a game. It can, at best, draw a game, but never score more points than its opponent. It goes for winning the war by losing battles.

That’s four traits:

Being nice
Punishing immediately
Forgiving immediately
Willing to lose battles

After publishing these results, and having learnt a lot about different strategies, Axelrod repeated the tournament. Four times as many entries poured in. This time from world experts on game theory. It also included an improved TIT FOR TAT called TIT FOR TWO TATS, which is an improved TIT FOR TAT that does not fall into a C - D - C - D cycle when playing against TIT FOR TAT.

TIT FOR TAT won again.

Till date, TIT FOR TAT remains an unbeaten individual strategy, and people believe it may be optimal.

(PS: By individual strategy, I mean that there are multiple programs that can team together, losing to each other and making sure that one of them wins. This sort of thing can beat TIT FOR TAT. But no individual program beats TIT FOR TAT.)

In our first term at IIMB, we played a game in our organisational behaviour class, intended to help us understand inter-departmental cooperation (or rivalry). The class was split into two ‘companies’. Each company had four divisions.

The game had 10 rounds. In each round, every division could choose to cooperate or defect. If everyone cooperated, each division made 3 points. If any division defected, it would make 5 points, while all cooperating divisions made 0 points. If all divisions defected, they would all make 1 point. The divisions were not allowed to talk to each other.

The aim was to beat the other company. (Not other divisions, within or outside the company.)

Our company started off with 3 Cs and a D, which quickly deteriorated to 1 C with 3 Ds by round 6. At round 7, it was 4 Ds.

Before round 8, we were all given a chance to have a huddle. A representative from each division would come together and talk things through. We promised to cooperate, and thereafter, it was 4 Cs to the end.

We lost the game. The other ‘company’ had started off with 1C and 3Ds, but had learned to cooperate pretty quickly, aided, in Prof N M Agrawal’s words, by “… Aparajita threatening the other divisions with her glares.”

The reason there’s a Prisoner’s Dilemma is the inability to reliably communicate or enforce behaviour. Having a chat helps. Having laws that punish you helps. Having a bully threaten you helps. The thing is, you need a signal of some kind. And it needs to be an early signal, or you end up waiting till round 8 and lose the game.

If you’re ever in a situation where cooperation helps everyone, but it’s not in interest to cooperate, here’s what seems to work:

See if you can agree to cooperate before-hand
- Have a chat
- Find policies that punishes defectors
- Threaten if required
If not, then try to force cooperation by signaling.
- Be nice
- Punish defection immediately
- Forgive repentance immediately
- Lose battles to win the war

Comments

Dibyo 19 Aug 2008 1:05 am: I really liked this piece. It’s not a new concept to me, and I’ve understood it well before, but the TIT-FOR-TAT part really connected.

Dibyo
(IIMB too, but a couple of years junior - we only heard about Stud Anand)
S Anand 19 Aug 2008 11:03 am: @Dibyo: Even when playing the game at IIMB, a few of us knew about TIT FOR TAT. Didn’t help. Guess there’s a huge gap between understanding and practice…
Abhi 28 Aug 2008 5:33 am: Fantastic piece, very lucidly put.

IITM - Mtech, we should have heard about Stud Anand.
deepak 30 May 2010 7:32 pm: We played the game in IIMB and ended up doing the same stuff and never getting it right. Dibyo, I think it was Prof. Reddy who made us play this game. But then I went to Solvay for Exchange - 2 each from IIMB and A , we tried TIT for TAT first( many teams too many combinations ), didnt help post the huddle while everyone else was trying for Tit for TAT we kept going against to drive home the message. The prof did mention that one enlightened party doesnt help unless everyone else gets the message :)
Prabu 10 Dec 2010 9:03 am: Hi Anand, Was going through your post. Just thought of this algorithm that would beat the TIT-FOR-TAT algo. it would be:
1. Never Co-operate. Now, this algo always either wins or draws individually. Group games are different. Thanks
S Anand 10 Dec 2010 2:56 pm: That’s an interesting point, Prabu: “group games are different”. Totally so. Your Never Cooperate algo, as you pointed out, always wins or draws. Tit-for-tat, interestingly, NEVER WINS AN INDIVIDUAL GAME. You can prove it, but this file is a demonstration: it plays random strategies out against Tit-for-tat. Tit for tat never wins. Maybe the lesson here is a stronger version of “Losing the battle, winning the war”. Perhaps it’s essential to lose the battle to win the war. (I’m probably oversimplifying, or in extrapolating analogies… anyway, just wanted to say: yes, never co-operate beats tit-for-tat on an individual game. Many other such strategies exist too. As you mentioned, group games are quite different)

Psychological framing in behavioral economics

Fri, 12 May 2006 12:00:00 +0000

People act on the spur of the moment, most of the time. So, as you can see from this example below, psychology can be more important than economics.

Senthil Mullainathan worked with a bank in South Africa that wanted to make more loans. A neoclassical economist would have offered simple counsel: lower the interest rate, and people will borrow more. Instead, the bank chose to investigate some contextual factors in the process of making its offer. It mailed letters to 70,000 previous borrowers saying, “Congratulations! You’re eligible for a special interest rate on a new loan.” But the interest rate was randomized on the letters: some got a low rate, others a high one. “It was done like a randomized clinical trial of a drug,” Mullainathan explains.

The bank also randomized several aspects of the letter. In one corner there was a photo-varied by gender and race-of a bank employee. Different types of tables, some simple, others complex, showed examples of loans. Some letters offered a chance to win a cell phone in a lottery if the customer came in to inquire about a loan. Some had deadlines. Randomizing these elements allowed Mullainathan to evaluate the effect of psychological factors as opposed to the things that economists care about, i.e., interest rates-and to quantify their effect on response in basis points.

“What we found stunned me,” he says. “We found that any one of these things had an effect equal to one to five percentage points of interest! A woman’s photo instead of a man’s increased demand among men by as much as dropping the interest rate five points!

Comments

Lesson 1: What Is Behavioral Economics? « The Innovative Economics Initiative 15 Nov 2011 6:25 am (pingback): […] (the background). This concept, presenting the same object in different formats, is known as framing, and is one of the most well-known biases used by Behavioral Economists. For example, if I asked […]
bala 8 May 2012 2:37 pm: i have written a tamil book on behavioral economics it can be downloaded from the link www.scribd.com/doc/88128740 the book குகைமனிதனும் கோடிரூபாயும் நூல் 1000 வாசகர்களுக்கு மேல் வாசிக்கப்பட்டுள்ளது may you go through it and give your comments. and forward to your friends if it is good thank you bala

Die later if it saves inheritance tax

Sat, 09 Mar 2002 12:00:00 +0000

Kopczuk and Slemrod demostrate that people will die later if it saves inheritance tax. I fully empathise.