Y. Wayne Wu

Product Research @Workato. HCI + AI background.

[Y. Wayne Wu @waynewu]

@Kay2289123 FREE is the biggest feature. Look at how much money you spending on AI every month. Muse enables millions of average people to do 80% of what you can do for free.

[signüll @signulll]

every major company is arriving at roughly the same offering:

persistent memory, email/calendar/messages, browser + computer use, background tasks, proactive notifications, voice, app/tool execution, ambient context, & some notion of a personal agent sitting above everything.

remarkable levels of convergence with very little differentiation whatsoever.

[Y. Wayne Wu @waynewu]

This is the craziest talent funnel most countries can only dream of. The brightest minds around the globe pay $$$ to enter this funnel, have to contribute to the economy year after year (through tuition or taxes), and only the top 30% who survive this brutal process get to reap the privilege of permanent residence. Getting rid of this process isn’t just killing the goose that lays the golden eggs. It’s burning down the entire goose farm.

[Jeremy Neufeld @JeremyLNeufeld]

The F-1 student to OPT pipeline is truly the "front door" for high-skilled immigration to the United States.

From: https://t.co/74b0lE8wKy

[Nicholas Fabiano, MD @NTFabiano]

Gaming may also just make you happier.

[Y. Wayne Wu @waynewu]

The chart is interesting but dated, so I used Astra to pull recent data and extend it using the author’s original methodology. It’s interesting to see that the “more time at home” pattern stabilize.
And WOW, what a time to be alive. This kind of follow-up research is just one prompt away now. Before AI, this would easily take me a few days of good work.

[Cole Jaczko @colejaczko]

is this genuinely the explanation for decline of everything - friendships, excitement, spontaneity, dating?

[Justin Skycak @justinskycak]

You cannot be creative at a high level unless you are robotic at a low level.

[Ethan Mollick @emollick]

AI seems to be beating the expectations of the best forecasters.

[Y. Wayne Wu @waynewu]

It’s wild to think most mathematical problems may remain unsolved simply because of high cost. Great mathematicians may act more like startup founders. Fundraising, storytelling, and selling the ROI of certain problems to the public.

[Abhishek Nagaraj @abhishekn]

By my calculations, just the compute here (300 BILLION output tokens) would cost a regular user $20-$30 million in tokens. How do we sustain this level of investment in science once the AI companies have decided that its no longer worth their time/effort to go after everything thats not a millenium problem.

also -- totally 🤯 - going to be hard to focus on another other research seminar/topic for a while.

[Yohan @yohaniddawela]

For six months, Google Maps sent a slice of drivers in ten American cities down deliberately slower routes.

About 30 seconds slower on average, and under 2% of trips were touched. The experiment ran on roughly 100 of the most congested road segments in each city, switching on and off day by day so each city acted as its own control. The results are out now in Nature Cities, from Google Research with collaborators at Berkeley and Stanford.

The reasoning behind it is old and slightly counterintuitive. Every navigation app does the same thing: it finds you the fastest route right now, for you alone. When millions of phones do that at once they all pour onto the same handful of arteries, and those arteries stop being fast. Transport economists have been writing about the gap between what's good for one driver and what's good for the network since the 1950s. Nobody had tested a fix at city scale with real drivers, because you'd need to steer routing for a large share of a city's traffic, and about three companies on Earth are in that position.

So Google put a penalty on the worst segments during their worst hours, which nudged the routing engine towards alternatives of similar road class and comparable travel time, then measured what happened across the whole network on weekdays between 7am and 8pm.

On the targeted segments, traffic moved roughly 2% faster in the median city. Los Angeles got 4.56%, Atlanta 3.30%. Fuel burn on those stretches dropped between 0.5% and 1%. Across every road that saw a change in traffic, which covers around 80% of each city's driving, speeds rose 0.35%, reaching 0.5% during the morning and evening peaks. Total travel time on affected trips fell 0.69%. Their Bayesian model put the probability that the speed effect was genuinely positive at 99.8%.

That adds up to more than 1,000 tonnes of CO2-equivalent saved per year, per city, in most of the places studied. Cars and vans account for around 10% of global carbon emissions, and the average driver spends about 2.6 years of their life behind the wheel, so a fraction of a percent across an entire road network is a serious amount of fuel.

Per driver, the saving comes to about 0.25% of an average journey, roughly one fortieth of the normal day-to-day wobble in how long the same commute takes. Nobody in those ten cities noticed anything, in either direction, including the people sent the long way round.

The authors are careful about what they haven't shown. They measured the immediate effect, not what happens once drivers work out the freed-up route is quicker and pile back onto it, which is the standard way road improvements get eaten. Their penalty scheme was deliberately crude, so the ceiling on the approach is unknown. And the underlying Google Maps data is commercially confidential, so nobody outside can check the numbers.

The finding sitting underneath all this is that a private company's routing algorithm has become a piece of transport infrastructure, tunable in the way a traffic light or a congestion charge is tunable. The obvious follow-up experiments are the ones a city government would want to run, not the ones a mapping company would.

link to full article: https://t.co/RlRlDLCSER

[Y. Wayne Wu @waynewu]

The wider distribution among AI users is really interesting. The key is to use AI to learn better, not to replace learning.

[Paul Novosad @paulnovosad]

Every college syllabus should include these graphs.

Use AI for homework, you will get it done faster and get a higher grade, and then get crushed on the exam.

[Daniel Litt @littmath]

Maybe too obvious to be worth saying, but: frontier models are now obviously superhuman at some mathematical tasks, including ones that the profession has, historically, rewarded with prestige etc.

[Ulugbek S. Kamilov @prof_kamilov]

Academia is fascinating. You spend years becoming the world’s expert on something eight people care about, and three of them disagree with you.

[Adam Shuaib @adamshuaib]

Genius is mostly an output game, not an intelligence game. These individuals often produced 10x the volume of their peers before striking gold.

We saw a neuroscientist with 300+ papers, of which 2 had made them famous, and a unicorn founder who had 20 unsuccessful products over nearly a decade before one was worth billions. We also heard the story of a showrunner with two Emmy-winning shows but pitches for 70 others unsold in a drawer.

The world sees the successes, but spouses/close friends are the ones who see hundreds of wreckages hidden behind the curtain. The world’s most successful people almost always have a long line of previous failures behind them if you dig deep enough.

[Y. Wayne Wu @waynewu]

@daniel_iversen @Workato @zapier Hi Daniel. Sorry to hear that and thanks for sharing your feedback. Do you mind hopping on a call to share your thoughts in more detail? We have a few features cooking that may help with what you’re experiencing. Thanks!

[Bilal Zuberi // Red Glass Ventures @bznotes]

I remind myself of this quite often.

[Alex Yablon @AlexYablon]

LLMs will always sound like millennials because no subsequent generation will be literate enough to produce much training text

[Nassim Nicholas Taleb @nntaleb]

@Jaffer22915438 The Roman army rule: action removes fear.

[Y. Wayne Wu @waynewu]

What this translates to in the AI age: Focus on a select few main things hard, and delegate everything else to agents.

[Dan Siroker @dsiroker]

One more “hack for sailing close to this wind”: having kids can force you to learn how to prioritize ruthlessly.

I’ve founded two companies that had successful exits. One before kids. One after.

With @Optimizely I made up for my lack of prioritization through sheer effort. I worked almost every waking hour of the day. Then when we sold I looked back (with the benefit of hindsight) and realized that much of that work didn’t ultimately matter in determining the outcome. Only about 5 things actually mattered. If I knew what those things were at the time I would have focused all my energy on them and stopped spending time on everything else. The challenge is you only know with certainty what those things are with hindsight.

With @LimitlessAI I was forced to take a different approach since I had three kids under 4 years old. I internalized the mantra: “the main thing is that the main thing stays the main thing.” I thought hard about what the main thing was at each stage in the company’s evolution and just focused on that (e.g. a key hire, the limiting factor on growth, a key engineering challenge), and didn’t spend time on everything else (e.g. networking, talking to junior VCs, worrying about competition, recurring 1:1s with direct reports). At 5pm every day I paused work to have dinner with my family and put my kids to bed. Giving myself that constraint meant I said no to a lot of things my former self would have done. Because of this constraint I ended up doing more impactful & productive work from 4:30pm to 5pm than my former self would have accomplished in a full day. Constraint breeds creativity. Embrace it.

[saila @sailaunderscore]

Berkson's paradox.

Tradeoffs don't have to exist within an entire population (and often don't) -- to exist within a selected portion of the population.

[Shreyas Gite @shreyasgite]

Karpathy manually went through every training image on ImageNet. Then the entire test set. Just to set a human benchmark.
He recently implemented a 1989 LeCun paper to measure how much efficiency you gain by simply knowing the deep learning future.
And these are just two examples we know about. There are probably 100+ of these curiosity-driven side quests.

Knowledge is not built by working at frontier lab or raising 100M but by the willingness to go absurdly deep on random questions just because you're curious.
My timeline is now full of people quoting and misinterpreting Karpathy. Maybe what they should actually take from him is this.