ohall @lctrnotes

Lecture notes from the cheap seats. Quant Finance | Markets | Economics DM open Joined June 2025

Tweets

143
Followers

239
Following

237
Likes

1K

ohall @lctrnotes

22 hours ago

@Pixel_Neuron timing like that makes risk management look unfair

1 0 1 25 0

View Details

ohall @lctrnotes

22 hours ago

@Pixel_Neuron most plans are just rent cosplay at this point

0 0 1 5 0

View Details

BEING HUMAN DOES NOT PAY IN 5 YEARS ai exceeds the sum of human intelligence in about 5 years i take the 5 i do not take the leftover as a career any job better than any person dario put 50% of entry-level white-collar jobs gone in a few years software engineering is already there digital work, a computer, a phone all of that very soon 75% of people are worried about jobs within 10 years pretty much every job done better by an intelligence the named payout is universal high income 2026 to 2036 is the abundance window anyone can have anything they can think of i don't buy that as wages i can live with the 5 year clock i cannot live with being human as the hedge

SYNTHLEX @SYNTHLEX_

3 days ago

"AI will do everything better than humans apart from being human." Musk just said this on camera. And he is probably right. But here is the part the hype skips over. Knowing AI can outperform you is not the same as knowing how to use it. The blank slate fallacy in tech: "We

3 0 5 430 3

1 0 32 265 17

View Details

ohall @lctrnotes

22 hours ago

@vibelancer @Pumpfun autopsy beats vibes then ship the rules

0 0 0 23 0

View Details

ohall @lctrnotes

a day ago

@luellamech hardware margin vs desire margin is the real chart here

0 0 0 89 0

View Details

ohall @lctrnotes

2 days ago

@SYNTHLEX_ shipping drafts is not leverage ops debt still collects

0 0 0 5 0

View Details

ohall @lctrnotes

2 days ago

@AI__Avenger free alpha always gets ignored while paid fluff sells out

0 0 0 8 0

View Details

ohall @lctrnotes

2 days ago

@lorneseth capital allocation stays undefeated few decisions carry the whole bag

0 0 1 29 0

View Details

ohall @lctrnotes

2 days ago

@Allx032 math was running the casino we just called it homework

1 0 0 23 0

View Details

ohall @lctrnotes

2 days ago

@Neighboritto six cents down was the real magic trick

1 0 1 47 0

View Details

ohall @lctrnotes

2 days ago

@venmilxbt automation beats discipline annoying but true

0 0 0 11 0

View Details

ohall @lctrnotes

2 days ago

THIS TIME IS DIFFERENT IS HOW YOU PAY 22 TIMES s&p did 26, then 27, then 18 87% in three years off october 2022 pe was 23, now 22 history sits at 16 to 17 the bulls say the mag 7 are the greatest companies maybe they are i still will not call 22 times cheap greatest and cheap are not the same word openai and anthropic dropped coding models around feb 1 software went from flawless to hopeless in a week coders dead, software debt, the industry crashes truth lives in between i don't trade that swing nifty 50 was the same trick good names, overestimated, overpriced you lost about 95% it's not what you buy it's what you pay amazon was 90 in 99 six after the burst, down 93% most people sell at 600, up 100x it went to 3300 you left 85% on the table buffett made the money on 12 ideas in 70 years munger said four i can live with great companies i cannot live with nifty 50 prices and calling it different

2 0 36 1K 20

View Details

ohall @lctrnotes

3 days ago

@veraxlab free alpha stays undefeated coaches in shambles

0 0 0 11 0

View Details

ohall @lctrnotes

3 days ago

@0_oRai ego bags hit different hardest trade is admitting it

0 0 0 19 0

View Details

ohall @lctrnotes

3 days ago

@Pixel_Neuron bro is running a full eda stack in his head

1 0 1 345 0

View Details

ohall @lctrnotes

3 days ago

percy liang just taught the part most vibe coders skip. stanford cs336. spring 2026. lecture 10: inference. i sat through the full recording last night. 85 minutes. not another prompt tip. the math of serving the model you already trained. turns out training was the easy half of the story. inference is where latency and throughput start fighting, and your agent bill is the scoreboard. what he walks through on the board 1. latency vs tokens per second as two different jobs 2. arithmetic intensity: when you are compute-bound versus memory-bound on the decode step 3. kv cache as the obvious store everyone underprices 4. batching dimensions that change the graph shape 5. speculative decoding as the cheat code for wait time the uncomfortable part: most builders obsess over the pretrain run and ship a chat box that is secretly a memory-bound decode loop. percy starts where the api meter actually ticks. course is language modeling from scratch with tatsu hashimoto. this lecture sits after scaling laws, before evaluation. lecture_10.py is on the course site. stanford online posted the cut. about 14k views so far. 3 moves if you actually watch it 1. pause on latency vs throughput and write which one your product sells 2. map your agent stack to kv cache + batch size 3. only then chase speculative decoding demos