Welcome to my homepage 👋 I miss the blue bird, so I made my own. No algorithm, ads, likes, retweets. No one else posting. Just me. Almost none of these were tweets. They're bits from my diary and notes over the years, things I found interesting enough to keep.

status: heads down, shipping small things. back online soon. #status

just like less context is better for some agents, it's sometimes better for the human mind to actively get rid of irrelevant info around to focus. aka context poisoning

randomness is uncompressible. a perfect compression scheme outputs something indistinguishable from coin flips, because any leftover pattern could be compressed further.

read this once and it changed my mindset: "if you die, your job will be posted before your obituary."

یارا بهشت صحبت یاران همدم است #سعدی

if i had to bet on someone, i'd take the person with a crappy draft who's mid-way through over the one with a beautiful book stuck on page 10. do crappy, but do it faster. quality comes with many many trials

smaller than your opponent in mount? stay chest to chest, knees tight, and when they bridge hard, redirect instead of resisting. their explosion becomes your transition. koala grip energy. #bjj

terence tao: if the theory has more params than the world it describes, it's not a good theory. it overfits.

asking a 1995 lawyer what share of legal docs will be word-processor generated: 100%. same with ai and code. the job stays: knowing what the software needs to do and having taste. not sure tho.

Patrick Collison's advice to 10-20 yrs: go deep on multiple things. pursue weird things that you want to pursue, not whatever the standard path is. heuristic: do your friends think your path is a bit strange? if not, maybe it's too normal. patrickcollison.com/advice

RAG cheat sheet: under ~500 pages, skip rag and stuff it all in context (cache the prompt). got keywords? use bm25. retrieve top 20 chunks, rerank, and have an llm add context to each chunk first. #ai

start simple. a single llm call with retrieval and examples is usually enough. add agentic complexity only when it demonstrably improves outcomes. #ai

there's a space between your feelings and the external things happening. pain is inevitable but suffering is optional. distractions/chaos are inevitable, just observe: notice it, gently bring your attention back. #focus

reflection. from 2 years ago: I wanted to learn #bjj, and I did I wanted to find a tech job, and I did I wanted to visit SE Asia, and I did I wanted to learn algorithms, and I did

be heavy. be close when you attack, close the gaps, flatten your opponent, take hard deep grips on limbs, use legs and core. #bjj

sweet spot for the mind: aroused, not stressed, not tired. mornings are for reading, never watching. If you think you perform great but nobody agrees with you, then you (by definition) are not doing great. #motivation

don't optimize a thing that should not even exist.

optimists get rich, pessimists sound smart. #motivation

a business doesn't fail at what you're great at. it fails at your weakest link, like sales if you're not a sales person. hire one more like your ace, don't accept mediocrity. #startups

Gradient descent is a blindfolded treasure hunt. you stand on a hillside, feel the ground for the steepest way down, and step. the hill is the loss landscape, the steps are parameter updates. #ai

maker schedule vs manager schedule: you need hours of focus to write code, managers live back to back in meetings. mornings code, afternoons meetings, or vice versa. focus is saying no #motivation

i built a chatbot over all lex fridman podcast episodes (1 to 325). ask a question, get an answer plus sources, and click to play the video from the exact moment it's discussed. #ai github.com/qniksefat/lexitalk

readme tip: gif > screenshot > text. and make the 'why' section a story: i had problem A, tried B, it failed because C, so i built D and now i do E with ease. #ai

Projections (Karl Deisseroth): watched the author's interview with Lex Fridman and decided damn this must be an amazing read as Lex enjoyed it this much. It wasn't and the training label back-propagates to Lex. could easily be divided into two books: one about the great facts and one with the stories too detailed. listened to the audible version where the author reads it himself. man are you this bored of reading your own book?!

Afraid of death? I have been dead before coming to life. Was I afraid?

empathy is NOT all you need. it actually seems to be an effect and not a cause of how kindly we treat peers. (Pinker)

to get unbiased data, find channels with a higher bar. If it’s instagram -> total BS YouTube a bit better A survey paper -> good A notebook on kaggle -> awesome The data itself! best. #ai

limbic capitalism: porn to sex, is what McDonald’s is to food

focus goes away. the hard part is gently bringing it back, and NOT rewarding the mind for being lazy. it's a muscle. accept boredom, you don't need to fill every moment with info. #motivation

better to have a product 100 people love than 100,000 who kinda like it. (yc startup school) #startups

too easy? make it harder. too hard? make it easier. and remember, lack of effort guarantees failure. #motivation

when you give up because you're tired, stressed or hungry: 1. you're not functional in that state because you never trained for it. 2. you're rewarding your brain for not working hard. #motivation

Guido van Rossum: ambiguity is a feature of natural language. like a dance. "I love you" is interpreted differently when you say it to your son or wife.

machine learning is the high-interest credit card of technical debt. quick wins aren't free: boundary erosion, entanglement, hidden feedback loops, undeclared consumers. (classic google paper) #ai

Grant Sanderson on learning: start with examples, then go to abstractions and formulas. do the problems at the end of the chapter. teaching sticks 9/10, seeing 5/10, listening 2/10, reading 1/10. #motivation

David Eagleman: brains are isolated in their own worlds, trying to connect with each other through sensors with a really low throughput pathway. #neuro

جریده رو که گذرگاه عافیت تنگ است. #حافظ

refactoring rule of three: first time, just get it done. second time something similar, do it anyway. third time, refactor. and it's like washing hands, keeps code cleaner.

Hippocampus is to neocortex what ram is to storage. non-overlapping patterns point to networks, like cpu registers in a von neumann machine. with hierarchical access you can address memories in log order. #neuro

سرود مجلس جمشید گفته‌اند این بود: که جام باده بیاور، که جم نخواهد ماند #حافظ

Dopamine is for #motivation and craving, more than pleasure. food, scrolling reels: you often find yourself not even enjoying it that much. (also a 10-30 min nap is magic)

جز یاد دوست هر چه کنی عمر ضایع است. #سعدی

غریب را دل سرگشته با وطن باشد. #حافظ

نظر به روی تو هر بامداد، نوروزی ست.

انگار که نیستی! چو هستی خوش باش! #خیام

آن‌ها که خوانده‌ام همه از یاد من برفت! الّا حدیث دوست که تکرار می‌کنم. #سعدی

از هرچه خورند کم شود جز غم تو #مولانا #rumi

خلوت‌گزیده را به تماشا چه حاجت است؟! #حافظ

چه خوش است بوی عشق از نفس نیازمندان #سعدی