1/5 In 2015, I won the ICPC World Finals as a member of the ITMO University team. It was the only time in Finals history when a team solved all the problems before the contest ended.
South Korean man Oh Yohan breaks the world record for the most pull-ups in a 24 hour period.
Insane.
Yohan did a remarkable 11,707 pull-ups in just 24 hours.
Yohan says he started training for the record during his military service back in 2019.
“Since my wife's second birth is not far away, I was hesitant to take on the challenge again and I was afraid to experience the extreme 24 hours again,” he said.
“But to give up like that, I felt like I would regret it for the rest of my life and it was too late to finish without taking on the challenge.”
Source: Guinness World Records
Sleep is the strongest medicine on Earth.
It burns visceral fat, repairs DNA damage & produces new brain cells.
And 2 out of 3 adults suffer from broken sleep...
Here's how you've been sabotaging it (& 6 foods that optimize it): 🧵
It’s 2012.
I sold my bootstrapped startup, made my 1st mil.
I wanna build a unicorn, raise from big VCs, move to SF.
One day I meet a guy looking like a movie star.
This day is gonna change my life.
He makes a pitch:
IMHO, Anthropic and OpenAI are playing a losing game trying to continue scaling compute by hemorrhaging equity capital losses (massively underwater CAC) as their core competitive strategy. Let me unpack this view:
Hyperscalers (specifically MSFT, META and GOOG) have orders of magnitude more resources (talent), users (billions in MAUs) and capital ($200B+ of working capital between them, to be exact.. and this could be 2-3X that quite easily with their liquid stock factored in) to build and scale SOTA models themselves. Additionally, they are in the lovely position to not need to monetize the models whatsoever - they can infuse them in their existing cash cow businesses and transform those value propositions for customers, boosting engagement, revenue expansion, data acquisition and a lot more. "How do we get customers to pay for the models??" is not even a question. You give the models away entirely. META does this explicitly with zero issue in the beautiful absence of a business model conflict.
As a result, I don't see a rosy future for Anthropic or OpenAI at all. Even in the enterprise market. They serve a useful purpose to give interesting product ideas to the hyperscale incumbents. And to entertain the industry. But not much else.
So... where is the real opportunity?
Radically more compute efficient approaches for achieving whatever approximates as "AGI" or "ASI"... which cannot be defined by anyone, but will indeed bring immense value for humanity. Some risks, but vastly more benefits than risks.
In the absence of massively (and I mean orders of magnitude) more compute efficient approaches being unlocked by a startup for scaling machine intelligence without training runs that cost tens of billions of dollars (which is the current regime in the next 6 months), the hyperscaler incumbents will absolutely take 90%+ of the market share.
There is no other path that I can realistically see happening.
gpt2-chatbot mysteriously dropped just 2 days ago.
People are doing wild things with it and can't figure out why it's so good.
9 examples (and how to access):