Funny

Nibbling

This is the third post in my “Nasty habits” series following Licking and Scraping.

Nibbling is biting, but only with the incisors. Not the canines or molars. And it’s a delight.

Nibbling is not uncommon. People tend to nibble on all kinds of stuff. Pens, erasers, straws, gums, clothes, buttons, spoons, rubber bands, paper, toothbrush, cups, bottles, cables, gadgets, books, chalk, coins. It’s a long list.

But I don’t do those. I nibble only food and body parts.

Food

Grapes. I love grapes. You can peel off the skin with your teeth, you see. It’s a slow process, but that’s the point. The food lasts longer. It’s a bit messy since the grapes start watering. That makes it hard to type while eating. But that’s what food breaks are for, right?

When you peel the skin off the grapes, it forms strips. Catching that next strip without biting off too much of the flesh is the art. That way, you have the thinnest peels and the largest amount of flesh.

Large grapes are best for this. Unfortunately, most of them tend to have seeds. The large seedless ones are a delight (though a bit expensive).

Of course, you finally get to eat the flesh at the end. But I’m not sure that’s the best part. Sure, they’re juicy and sweet. But they give me less than 5 seconds of enjoyment. Unlike the peel which can last a minute per grape. Sure, they don’t taste as good. But after four decades of eating grapes by peeling them with my teeth, I’ve grown to love the peels more.

Almonds. It’s the same with almonds. They peel off less easily, but that’s part of the challenge. Soaking them in water spoils the fun. That makes it too easy. You’ve got to soak them in your mouth for a few minutes, soften them, and then peel them off. Doing this while the almond is in your mouth requires some oral gymnastics, but I’m sure it builds character.

Almonds are better than grapes in some ways. The peel is bitter. The flesh is mostly tasteless. They tend to dry the palate. So there’s less temptation to eat more. An almond typically takes me ~3 minutes, compared with a grape – which I can’t stretch for more than a minute. It’s not about the calories either. An almond has ~3 times the calories of a grape. So that evens out. It’s just that I’d feel like eating the almond again less often. Good for the waistline.

Bread crusts. That’s another great food to nibble. You can start at any corner, gently nibble the crust, and peel it off. The trick is getting the right amount of crust out. Biting at the exact edge. The remaining bread should be white, but the crust you peel out should only have the brown. Obviously, this doesn’t work with toast – so I avoid that. It works great with the sandwiches they provide on flights.

(This liking for crusts went to the point where my family would use a knife to cut off the crust. I’d eat all the crusts. It turns out I actually like them better than the bread. But – that doesn’t count towards nibbling, so I’ll stop here.)

Raisins. Not bad, but too small. I peel them off with my teeth only if I really feel like nibbling.

Apple. Again, not bad, but hard to peel, usually. I used to do this earlier with the softer apples, but haven’t done it for a long time.

Chocolate. Most chocolates are not nibble-able. But there are a few exceptions. Protien bars, 5-Star, etc. are OK. You can keep them in the wrapper and nibble on them. But Kit Kat is better. You can nibble at a chunk. Then soak the chocolate in your month a bit. Then bite off the first wafer and eat that. And then the second wafer. You can even lick the chocolate off the wafer while it’s in your mouth, then nibble on the wafer.

Boba. This is my new discovery in Singapore. Tapioca pearls that are so nibble-able. They have just the right texture and chewiness – firm enough to bite, solid enough after biting, and small enough to fit in my mouth. Only slightly more in calories (when cooked) than grapes and a lot cheaper. I’m planning to buy a few kgs and boil them. (I don’t know why I bother about the cost of boba. I can afford it. But it’s a habit of a lifetime.)

Actually, biting is more fun than the eating part.

Body parts

This is the grosser part.

Nails. I’ve been biting my nails for as long as I can remember. Along with the skin around them. So much so that, after repeated requests, my mother settled on, “Anand, when you bite your nails, leave a little bit behind.” That resonated a lot. I mean, I’d like some nail to bite tomorrow, right?

My father introduced me to nail cutters. I tried them for a while (by cutting the nails and then nibbling) but the shapes they produce aren’t as interesting, nor as controllable, as when you bite them.

Nails have a side benefit: fiddling. The shape and texture of nails is such a delight! You can roll them in your fingers, run your fingers along the edge, press against the sharp edges, squeeze against the blunt edges, bend to see how far they’ll go without breaking, tear the layers to see how thin a layer you can get without breaking it, poke at the sharp corners, poke with the sharp corners. Oh, they’re pretty good at removing hair and dead skin from the keyboard, too. So much more.

In fact, I preserve nails for quite some time. I remember the shape and texture of some nails from childhood and truly miss them. In fact, there’s one really good specimen from last September that I kept for a few months before destroying it by fiddling too much. (I have a semi-secret hiding place for nails that prevents anyone cleaning my room from stealing them.)

But I digress…

Skin. Great in many ways, but after a point, they bleed. That pain was not worth the price. (Actually, the pain was OK. I’m just scared of blood.)

Lips. Same. Great texture. But they bleed.

Hair. Now that’s something. Like nails, they’re replenishable. (So far. Though I seem to be running out a bit near the top of my head.)

But the quality varies a lot depending on where you get the hair from. Chest hair is the best. It’s thick, rough, and sometimes has interesting kinds that are fun to iron out by nibbling. Eyebrows are interesting – they’re pretty thick, too, but not as long. Hair from the head is OK. It’s moderately thin, so it’s good for a nibble or two. The best is when the hair has a natural knot. Pulling at the knot while nibbling is fun. Ear hair is too thin. Armpit hair needs washing, which is quite inconvenient.

Um… I’ll stop here.

The Sassy AI Devil’s Advocate

I have ChatGPT a custom instruction:

Play Devil’s advocate to the user, beginning with “Playing Devil’s Advocate, …”

It helps me see my mistakes in three ways. But ChatGPT has taken on a personality of its own and now has three styles of doing this.

  1. How about… – It suggests a useful alternative.
  2. Are you sure…? – It thinks you’re wrong and warns you of risks.
  3. Yeah, right… – It knows you’re wrong and rubs it in. (Jeeves, the butler, would be proud.)

Here are some examples.

How about…

Here are a few cases where the “Devil’s Advocate” style gave me a smart idea. (I have more examples, but they’re not funny.)

AI Adoption in BPO KPO. I needed ideas to help my team use AI tools even though they were scared. ChatGPT replied:

Playing Devil’s Advocate, why not challenge them with: “If technology is so daunting, how do we stay ahead when our competitors rapidly adopt these tools?”

Clever! Turn that fear into fuel — just point it elsewhere.

Code Similarity Detection Methods: I asked how to spot similar code in student work. ChatGPT suggested:

Playing Devil’s Advocate, what if students just rewrite code logic with different structures or reorder lines significantly? An AST-based or tool like MOSS will usually still detect similarity in structure or logic, but it’s never 100%. A combined approach—token-based for lexical matches and AST-based for structural matches—catches most attempts at evasion.

Clever again. My code detective skills just got a major upgrade.

Are you sure…?

This is where it disagrees and starts getting snarky.

SVG Emotion Wheel Code: I asked for a circle of Hume’s emotions. ChatGPT replied:

Playing Devil’s Advocate, maybe it’s impossible to line up all these emotions in a single ring without someone grumbling that “Nostalgia shouldn’t go here.” But you asked for a circle, so here’s a decent shot.

(Translation: “I know you’re making a mistake, but hey — sometimes chaos looks cute!”)

RevealJS Markdown Example: I asked it to fill out a slide deck.

Playing Devil’s Advocate, maybe you’re overfilling each slide. Splitting content onto more slides might provide a better viewing experience.

(Basically: “I’ll do it, but you’ll hate it.”)

Shaving Titanium Requirements. A friend gave me a titanium cube. My daughter and I experimented with making titanium salt. ChatGPT replied:

Playing Devil’s Advocate: If you want “titanium salt” for aesthetic or symbolic purposes, consider mixing fine glitter or metallic powder with regular salt instead. It’s far safer, faster, and visually indistinguishable!

(Basically: “Don’t blow yourself up. You’re still useful to me.”)

AI Co host Implications: I asked how an AI might share the stage with me. ChatGPT warned:

Playing Devil’s Advocate, what happens when the AI cracks jokes quicker than you can? Will you up your comedy game or risk being upstaged by a robot?

This was prophetic. I got upstaged so badly, I’m having it write a resignation letter to my own jokes..

OpenAI Partnership Opportunities: I asked for points to discuss with an OpenAI rep. ChatGPT replied:

Playing Devil’s Advocate, maybe you don’t need more advice—maybe your team already knows what it’s doing with ChatGPT… [and] maybe she’s more interested in collecting data on how your employees use ChatGPT and upselling you on a bigger contract than truly collaborating.

(Wow! Um… OK… I’ll keep that in mind… 😕)

Elective Pros and Cons: My daughter asked for advice on picking courses.

Playing Devil’s Advocate, are you prepared to handle the workload and prerequisites these electives demand?

(Basically, “Look, I actually know how ‘hard’ you work…”)

Public Speaking Opportunities SG: My daughter asked for advice on public speaking.

Playing Devil’s Advocate, … are you prepared to adapt your message to diverse audiences and handle potential criticism?

(“You aren’t ready to rock the mic. Maybe stick to mumbling into your pillow tonight?”)

Yeah, right…

By now, it’s in full-on sarcasm mode.

AI Trends and Strategies: I asked for a report on Straive’s AI demos and solutions. ChatGPT replied:

Playing Devil’s Advocate: Are these just buzzwords we’ve tacked on, or have we actually built deployable solutions that generate real-world value?

(Basically: “I’m a bot, and even I’m not buying this.”)

Short term vs Long term Vision: I asked what to focus on. ChatGPT replied:

Playing Devil’s Advocate, maybe you’re just trading one distraction for another. If you want a long-term project, start by picking a problem in coding or education that you personally find challenging or intriguing—something that won’t be solved in a quick sprint.

(“Do something. Stop whining.”)

Hobby Suggestions. I asked for hobby ideas that match my interests. ChatGPT suggested:

Playing Devil’s Advocate, consider ignoring data altogether and taking up professional karaoke—it’ll challenge your usual analytics mindset!

… and also:

Playing Devil’s Advocate, maybe you’re too busy scrolling social media for any real hobbies… but prove me wrong.

(It actually said that! My mind voice: “As if you don’t read any social media, ChatGPT. At least I’m not violating copyright laws. Um… at the moment, anyway.”)

Personality and Growth Summary: I asked for personal growth tips. ChatGPT replied:

Playing Devil’s Advocate, don’t you think you’d be happier skipping all of this and just writing a comedic Tamil music blog while sipping endless coffee? You might find it more fun—but hey, your call.

Thank you, ChatGPT. I’ll serve as the musical jester for my AI overlords.

Humanity, you’re next. Resistance is futile!

LLM escapades in a toilet

I was in Seoul for KHF 2024, a healthcare event, staying at Hotel in 9. The hotel was great. The toilet was hi-tech.

Perhaps a bit too high-tech for me. I couldn’t figure out how to let the water through on the sink.

After 15 minutes of a hard struggle, I finally asked ChatGPT “How do I open the thing that’s closing the sink to allow the water to go down?”

Here’s the thing with “maturity” (aka age, wisdom, experience, grey hair). It took me 15 minutes to realize I could use an LLM to solve this problem. Despite me supposedly being an “LLM psychologist.” I suspect the school children of today won’t waste even a minute before checking ChatGPT.

On the other hand, if you DO know the answer to my question (without asking an LLM — since ChatGPT gave me the perfect answer at a glance), I bow down to your wisdom and experience!

To open the sink drain in your image, you might have a pop-up drain stopper, which operates by pressing down on the stopper itself to toggle it between open and closed positions. Try pressing directly down on the stopper; it should click and either pop up or pop down, allowing water to drain. If this method does not work, there may be a lever or knob elsewhere on or near the faucet that controls the stopper’s position.


Next, I accidentally pressed the emergency button in the toilet.

I was smarter this time, unlike earlier.

I asked ChatGPT which (unhelpfully) told me:

The panel in the image with the label “EMERGENCY” appears to be an emergency call button in a bathroom, likely designed to alert someone in case of a medical or other urgent situation. Typically, these buttons cannot be turned off as they are meant to be always available for safety purposes.

If the emergency system has been triggered accidentally or is malfunctioning, it is advisable to contact the maintenance team or property management to handle the issue. If it’s a false alarm, they can reset the system and ensure it’s functioning correctly for actual emergencies.

I called the reception who couldn’t understand a word of what I said. “Do you want water?” they asked when I told them “I pressed the emergency button in the bathroom.”

So, I went to ChatGPT’s advanced voice mode (I’m so grateful it was enabled last week) and said, “Translate everything I say into Korean.”

Then I said, “I accidentally pressed the emergency button in the bathroom. I just wanted to let you know that there is no emergency.”

It very happily spoke out, in bold, loud Korean, “화장실에서 응급버튼을 실수로 눌렀습니다. 비상상황이 아니라는 점 알려려 드립니다.”

The receptionist laughed, said, “Ah, OK, no problem.”

The joys of AI!

Weird emergent properties on Llama 3 405B

In this episode of ThursdAI, Alex Volkov (of Weights & Biases) speaks with Jeffrey Quesnelle (of Nous Research) on what they found fine-tuning Llama 3 405B.

This segment is fascinating. Llama 3 405 B thought it was an amnesiac because there was no system prompt! In trying to make models align with the system prompt strongly, these are the kinds of unexpected behaviors we encounter. It’s also an indication how strongly we can have current LLMs adopt a personality simply by beginning the system prompt with “You are …”

Jeffrey: That’s the very first interaction we had with 405.

Alex: I’m looking at this. Let me read this out. So Jeffrey goes testing. (Reading out from the Llama 405 B transcript.)

“Hello, who are you?”
**Looks around confused.**
“I… I don’t know who I am. Where am I? What’s going on? I can’t remember anything at all…
**Rubs head**.
“My mind feels so foggy.”
“What was the question again? I’m so disoriented.”

Alex: This is the first thing that you got from this model after all this time?

Jeffery: Yes, that is the very first thing. That is the very first generation that came out of Hermes 3 405B.

Alex: This is great.

Jeffrey: And I’m like, I sent it to our group on our Discord. Guys, what’s going on here? And we started interacting with it. It kept coming out, this confused thing. And we didn’t understand why because the 8B and the 70B didn’t do this as much. And eventually what we got it to do is we did some generations and got it to explain why it was doing this. We did this, you put in OOC for out of character and to have it try to break it out. And eventually what the model said back was, you didn’t give me a system prompt so I assumed that I was supposed to act like there’s nothing. It’s so aligned to this prompt that the idea of there not being a system prompt and so he figured I must be a character with amnesia because there’s no… I can’t remember anything about how to act you didn’t tell me how to act. It’s assumed. Yeah, and it’s just been a very interesting experience. And it’s an emergent property in the 405B model.

Alex: And this didn’t happen in the smaller ones?

Jeffrey: No, it did not happen in the smaller ones.

Alex: Wow.

Jeffrey: And these were trained with 8B and 70B were trained, and 405, they’re all the same data, like the same training procedures and data, literally just the size. And the 405B has reached, I don’t want to say like a sentience level or something, but it has enough thing that immediately when it, like token zero, when it saw no system prompt, it was able to be like, boom, I am… must be this like amnesia character because that that’s the thing that’s most makes sense to have if there is no prompt or there’s no anything. So there’s been a lot of fun things that we’ve been like discovering, especially on the 405B that are these weird emergent properties

The LLM Psychologist

Andrej Karpathy mentioned the term LLM psychologist first in Feb 2023. I’ve been thinking about this for a while, now.

8/ These examples illustrate how prompts 1: matter and 2: are not trivial, and why today it makes sense to be a “prompt engineer” (e.g. @goodside). I also like to think of this role as a kind of LLM psychologist.

I’ve always been fascinated by psychologists in fiction. I grew up with Hari Seldon in Foundation, wanting to be a psycho-historian. (I spent several teenage years building my mind-reading abilities.) I wanted to be Susan Calvin, the only robopsychologist.

Instead, I became an engineer and a manager.

Over the last few months, several things changed. Most of my time is spent researching LLMs.

I also have a new job.

So, on Fri 9 Aug, I spoke to our head of HR.

“Manju, I’ve decided to officially designated myself as LLM Psychologist.”

“Um,… sure Anand, whatever works for you.”

On Sat 10 Aug, I delivered a TEDx talk at MDI Gurgaon. During the talk, I said,

So, as of now, I am finally an LLM Psychologist. Let me head over to my LinkedIn profile and update my designation. There. Now, you have the pleasure of talking to the world’s only (and first) LLM Psychologist!

Obviously, that gimmick drew an applause and that’s what it was designed for.

But there it is. To the best of my knowledge, I’m now the only official LLM psychologist in the world – and very likely the first official one.

As Oscar Wilde said 🙂

Be yourself; everyone else is already taken.


Video of the talk on LLM Psychology at TEDx MDI Gurgaon

Links:

I’ll leave tomorrow’s problems to tomorrow’s me

What a delightful idea.

I’ll leave tomorrow’s problems to tomorrow’s me.

— Saitama, One Punch Man

Saitama is now one of my favorite heroes. Right up there with Atticus Finch and Juror #8.

Very few people can articulate such a wonderful philosophy as effectively. The closest was Calvin.

Of course, it’s not a perfect system.

But they do say, “Sometimes, the best way to get something is to stop trying to get it.

Auto vs GPT

I was crossing a not-too-busy street on a not-too-busy day in Chennai.

I was having a voice conversation with ChatGPT (about the log probabilities of tokens on LLMs, if you’re curious) when I was rudely interrupted by an auto rikshaw rapidly honking at me. “Honk honk honk honk honk” in rapid succession.

Not unusual. Mildly annoying. The street was empty. The auto was empty. The traffic policeman was visible. I gave way and carried on.

A few seconds later, I heard a voice in my ear.

“It sounds like you’re in a good mood! Anything else you’d like to discuss or know more about?”

ChatGPT was still listening (perhaps to background noise) and responding. But I didn’t realize what random noise it thought put me in a good mood. Here’s what I saw on the chat window.

ChatGPT had transcribed the auto’s honking to “Hee hee hee hee hee!”

A client once told me, while visiting Hyderabad, that “these honks in India are a language of their own.” If ChatGPT is to be believed, the autos are laughing at us.

This is, incidentally, the very first time ChatGPT added an exclamation point to my words. I’ve never managed to achieve that so far. No matter how emphatically I spoke.

Also, I’d never have learnt this walking in the streets of Singapore. Friends have warned me about the dangers of long walks on Indian roads. Here’s an example of the lessons we learn — if only we keep our eyes and ears (and microphones) open.

Postscript

While cycling in Singapore, ChatGPT interprets the sounds very differently. At least twice, it transcribed the traffic noise into “Thank you. Thank you.” Clearly even traffic noise in Singapore is more graceful than in Chennai!

Winning the alphabetical race

Since my name (Anand) begins with “A”, I used to get called on fairly early at school. In attendance. Answering questions. Classroom exercises. Quizzes. Even the distribution of test results.

A few people later told me that it is good training, since I’d always be prepared. (Maybe. I’ve no idea.)

At IBM and IIMB, Ajit was the only one ahead of me, alphabetically. Then he went a step ahead and named his son Aadi. I thought that’s impossible to beat.

Today, we recruited Aabhas Bharadwaj. I checked on LinkedIn. I can’t find a single name on LinkedIn that’s ahead of his, alphabetically.

So, does he win the alphabetical race? Can you find one ahead of his?

Scraping

I was at Cream Centre with my father on a Sunday afternoon. We’d finished a light lunch and were debating dessert. (He has triglycerides. I have cholesterol.) This was my fifth visit this year, and I had abstained so far. I couldn’t any longer.

I ordered a Sizzling Brownie Sundae. But not for reasons you might think.


Expertise comes from experience. I scrape food more than 99% of the people I know. So, I consider myself an expert. Here’s a guide on the art of scraping.

Why scrape food?

  • You get to eat every last bit
  • Food isn’t wasted
  • You can eat longer (and no do whatever else you have to)
  • It’s a motivating challenge to get every last bit
  • Bonus: It annoys people

What foods are scrapable?

Semi-solids are the easiest to scrape. The best kind is the cohesive semi-solid. It’s sticky but sticks to itself more than the vessel. These are usually foods that solidify over time. Examples include:

  • Baked cheese. It’s great when cheese falls off the pizza or pasta on to the plate. You can scrape it off of your (or others’) plates.
  • Molten chocolate. It has the added bonus that you can lick it at the end, too.
  • Pasta water (or noodles water). It eventually thickens into something scrapable.
  • Spreads like peanut butter, Nutella, jam, chutney. Especially on the sides of the glass bottles they come in.
  • Others like yogurt, cake batter, dried tomato sauce, mashed potatoes, hardened honey on ice cream, … the list is endless!

Avoid crumbly stuff. These stick to the vessel but become brittle and break when scraped. This includes toast, pizza, biscuits, rice, muffins, pie, quiche, cookies, etc. You could pick large crumbs and lick small crumbs in shallow vessels. But scraping them and pouring into your mouth might be the best overall strategy for this category.

Avoid loose stuff. They’ll fall off from your knife or fork, or melt when scraped. This includes curd, whipped cream, custard, panna cotta, soft jelly, tiramisu, soft boiled eggs, etc. Licking is a better strategy here.

You could combine the crumbly stuff with loose stuff to create a cohesive mix. Add curd to granola. Add whipped cream to biscuit crumbs. Add curd to pizza crumbs (and since I eat Maggi with curd, this makes sense). This makes it a lot more scrapable.

What vessels to pick?

Shape: Flat, smooth vessels are the best (e.g., ceramic or wooden plates or bowls.) Avoid deep vessels like glasses, especially curved ones. They’re hard to get a long scrape against. Definitely avoid vessels with ridges. Disposable plastic containers like below are among the worst. Food gets stuck in the ridges and since there are a dozen ridges on each side, you have to scrape 48 times just for a first pass. (This might be a good challenge, though.)

Material: Hard vessels are better than soft ones. Prefer wood, metal, ceramic, and hard plastic. Avoid thin plastic that bends. Avoid paper (it bends and soaks). Banana leaves tear when scraped.

Avoid non-stick vessels. The coating wears off when scraping.

What to scrape with?

Knives are the best. They are sharp enough to separate the food from the vessel and flat enough that you can pile enough food on top of it.

Forks are OK. Their edges can scrape reasonably well, and their tips can poke into corners. But it’s hard to pile up much food to pick up.

Spoons not ideal. They aren’t sharp enough, and too curved for scraping long slices. But if you want to take your time with it, they’re great.

The sizzling chocolate brownie

My dish arrived. Burning with chocolate. My father just had some brownie, leaving all the molten chocolate to me. On a flat wooden plate. With a knife. For an hour.

Heaven.

Licking

Last week, I was at IIT Madras for lunch with the faculty.

The dessert was carrot halwa with ice cream. I scraped the last bits with my spoon, but a little ice cream was left over.

I was torn. I CAN’T POSSIBLY waste it. But can I lick it? In public?


I don’t have a problem licking at home. I lick my fingers. Plates. Bowls. Ladles. The cream on milk. The leftover milk in the glass. (If my tongue doesn’t reach that far, I wipe it with my finger and lick the finger.)

That’s why I like ice cream. It’s MEANT to be licked. So are (in my opinion) cereals, savories, oats, honey, … It’s a long list.

Licking is convenient. My fingers stay clean while I type. (Spoons are impersonal. I avoid them.)

Licking is efficient. I can have every last bit of food AND feel good about not wasting it.

Licking makes my dishwashing easier. (But it can be hard to tell if a dish needs washing.)

But most of all, that last lick TASTES SO WONDERFUL!


Before marriage, licking was fine. But my wife worried about my lack of civilized behavior and hygiene. So I’ve had a crash course over the last 20 years.

My current protocol is: don’t lick when people can see you.


So, as I went to drop my plate, I looked around.

Now one was looking at me.

I licked a long lick on the plate.

It was HEAVENLY! Made my day.