Eason
100 posts
@kylegawley Everytime I hear an infra engineer's pager go off I think to myself: "Value to the customer"
English
I quit Notion
vibe-coded my own version in 3 hours
crashes every 20 minutes
burns $200/mo in server costs
but at least I own my data
English
Jensen name-dropped me in the keynote and posed with our belt.
He has a physical belt too but they just showed the pic
Intially I made fun of the 35X perf improvement being bogus, I thought it was an exaggeration of performance
Turns out he was sandbagging, and perf is 50x
English
my Taiwanese immigrant mom begged, borrowed and stole to send her 4 kids to private school. All 4 of us got kicked out but she'd do it again tomorrow.
Heidi N. Moore@moorehn
This is a common, very bourgeois error about the purpose of private school. The purpose of private school is not academic achievement. It is to establish a child within a social network of contacts that will help their career and marriage prospects for life.
English
This implies they need to be paid more
adel 🌟@adelwu_
everyone wants to hire fdes but nobody wants to be one
English
@natolambert He probably has the most to lose from open models disappearing. 1000% though Nvidia is the one place with enough capital to push this ecosystem along. I hope they achieve Chrome levels of domination here.
English
Jensen on Nemotron 3 Ultra and it's role in the open ecosystem. It's great to have Nvidia leading on this, with so much turnover in open model providers in recent years. Super excited to see how the model turns out.
English
the learning algorithm for AI is incredibly elegant and simple, but its rewards are messy and complex
likely: the learning algorithm of the brain is incredibly elegant and simple, the rewards are messy and complex
English
@NVIDIAAIDev All I could think of was @dylan522p's shiteating grin when this happened 🤣 congrats y'all
English
@natolambert @karpathy i was amazed to see the new updates hes still making to nanochat and other teaching tools its true GOAT behavior
English
I personally think about this a lot. We all have a huge desire to be at one of the 3 companies at the front edge of AI, but the ecosystem can't work without independent voices guiding and understanding progress.
@karpathy is the GOAT at this. It's a different path to impact.
Noam Brown@polynoamial
@saranormous @karpathy @NoPriorsPod Why is he not at a frontier AI lab at the most pivotal time in human history since at least the industrial revolution?
English
@_dollgirls The japanese do this too. Tokyo 東京 just means east capital. The other cardinal direction 京s are all in china. (北京beijing) north capital, （南京nanjing) south capital, (西安) peaceful west but it has been callled (西京 west capital) before
English
@jasminewsun @charlesxjyang @jasoncrawford @Steve_Yegge @zebriez @kevinroose I literally framed why I was applying to a lab based off which dictatorship was most benevolent. Felt pretty depressed after that call.
English
@charlesxjyang @jasoncrawford @Steve_Yegge @zebriez "OpenAI vs. Anthropic product culture" and "how DeepMind turned around Google's AI efforts" will be in the forthcoming @kevinroose book!
personally my glib shorthand is that Anthropic = benevolent dictatorship, OpenAI = too Hayekian, and GDM (at least pre-merge) = Kafkaesque
English
This kind of writing is particularly lacking for tech companies
Pieces like @jasoncrawford on Amazon's 2-pizza teams, @Steve_Yegge's Google platform rant, @zebriez on Stripe's written culture are too rare and far between
So many stories waiting to be written: OpenAI vs. Anthropic product culture, why Siri failed, how DeepMind turned around Google's AI efforts — these are all stories about corporate bureaucracy, not just technology
roon@tszzl
much of the nature of the world is explained by the nature of bureaucracy and yet - bureaucracy is rarely written about well beyond cliches. the great authors and people who worked in a large organization are generally disjoint. more common in east asian media
English
@natolambert I think the singularity narrative got a lot more gas from RL being able to leverage compute to increase capabilities and not be data constrained.
There's real and amazing advancements from all this RL but it still looks like it has a limit in total compute available.
English
People overuse the singularity as the term to describe the large acceleration in ai progress that'll come due to better agents. There are still real points of friction that the current models don't address. The singularity is catchy but misleading. Progress isn't infinite.
English
just read a reddit post and a youtube breakdown of the Anthropic SWE interview loop
never felt more dumb in my life.
apparently you need to casually:
- implement LRU cache with concurrency
- build a web crawler
- design LLM inference infra, KV cache management
- reconstruct profiler traces from sampling profiler
AI companies are interviewing like this now.
English
@bcherny curious why this isn't enabled by default? I'm assuming the CC team evaled this internally and decided against it as an overt decision
Om Patel@om_patel5
claude code has a hidden setting that makes it 600x faster and almost nobody knows about it by default it uses text grep to find functions. it doesn't understand your code at all. that's why it takes 30-60 seconds and sometimes returns the wrong file there's a flag called ENABLE_LSP_TOOL that connects it to language servers. same tech that powers vscode's ctrl+click to jump straight to the definition after enabling it: > "add a stripe webhook to my payments page" - claude finds your existing payment logic in 50ms instead of grepping through hundreds of files > "fix the auth bug on my dashboard" - traces the actual call hierarchy instead of guessing which file handles auth > after every edit it auto-catches type errors immediately instead of you finding them 10 prompts later also saves tokens because claude stops wasting context searching for the wrong files 2 minute setup and it works for 11 languages
English
@chongz @alexandr_wang A lotta folks just wana see him fail. Even I feel a unfair bias against Alex driven by the narrative that Scale was just "cheap human labor outsourcing aaS".
But I decided to just shut up and hope MSL makes something 🌶️. It's on Zuck's dime anyway.
English
Seeing a good amount of folks quote tweeting this to highlight and affirm their priors around Meta and @alexandr_wang
Guys this article is armchair speculation from a Jakarta publication quoting the Times of India 🤦 We're so cooked once AI generated slop becomes more refined.
Piotr Pomorski@PtrPomorski
No way, a product that sucks doesn’t work as expected! idnfinancials.com/news/61918/zuc…
English
Terrible zero aura rebrand
Brian Hioe 丘琦欣@brianhioe
The KMT has done a drastic rebrand on visual identity lately. Er, take a look for yourself
Español
@dylan522p I remember when this was the most well known bit of Notion's culture but never got to experience it myself.
English
The number 1 reason Cursor and Thinking Machines should merge is not because one has amazing product and the other research.
It's because they have strong culture fits, they both are no shoes in the office.
English
When we turn the clocks ahead on March 8, it will be the last time change ever for BC. We're changing to a permanent daylight saving time, simply called Pacific Time (PT).
Learn more: News.gov.bc.ca/33415
English
@GergelyOrosz @bcherny I remember the days where capacity planning was just ordering 2x the hardware you needed and having live servers running at <30% utilization
aka four years ago.
English
@bcherny yes except AFAIK we could source hardware ahead of time and plan capacity for the inevitable 10x growth :)
platform reliability was higher than this and a massive focus - it needed to be!
but it was not as rapid growth as now, also true, and more predictable to prep for
English
On one end, the Anthropic team is a massive user of AI to write code (80%+ of all code deployed is written by Claude Code). They ship amazingly fast.
On the other hand, seeing these beyond terrible reliability numbers suggests there might be a downside to all this speed:
English