HNHacker News
TopNewBestAskShowJobs

emmett

4,858 karma · joined February 19, 2007

Founder and CEO of Softmax. Softmax is building massively multiplayer benchmarks for coding agents to train social intelligence.

Previously very interim CEO of OpenAI, YC Partner, and founder and CEO of Twitch.

Email me at emmett@softmax.com.

submissionscomments
emmett··on Recurrent Net Dreams Up Fake Chinese Characters in Vector Format with TensorFlow
Of course it's prescriptive and there are regional differences. As you say, it's just like spelling. "Colour" still looks wrong to me. Just like a different stroke order looks wrong to a native reader. The point is the differences are visible to the reader, it's not just an arbitrary set of rules for writers. The reader can tell if you're following the rules or not.
emmett··on Recurrent Net Dreams Up Fake Chinese Characters in Vector Format with TensorFlow
I think a key point about this, which was only clear to me after I learned to write Hanzi in Mandarin, is that if you write the strokes in the wrong order it will not look right at the end.

To someone looking at characters for the first time, it's not obvious that the stroke order matters to the reader as spelling does in English. It's not about fetishization of the process over stroke order, it's really obvious to the reader if you wrote the character incorrectly.

emmett··on Firms help websites evade ad blockers
You've seen ads, unless you've managed to avoid almost all video content. They're just "native ads" -- advertisements worked into the content itself. It's just another evolutionary arms race...
emmett··on Why Homejoy Failed
And now I am sad again because cleaning service discovery is still broken. Not that you're wrong, just that it would be really nice to be able to find cleaners through a service with good reputation tracking and recommendation. Which is intrinsically difficult to build for exactly the reason you specify.
emmett··on An inside look at YouTube’s new ad-free subscription service
“The children now love luxury. They have bad manners, contempt for authority; they show disrespect for elders and love chatter in place of exercise.”
emmett··on A Former Twitch Employee Has One of the Most Reproduced Faces
Technically he uploaded the picture himself when he was working at Twitch (then Justin.tv) -- it used to be an employe e benefit that you could add your face to the emoticon list.

As to how Josh feels about it, I leave you with this link to my TwitchCon keynote presentation: http://www.twitch.tv/twitch/v/17714623?t=37m0s

emmett··on New Findings from War on Poverty: Just Give Cash
[citation needed]

What makes you believe this is true, beyond simple assertion?

emmett··on Yale to Be Paid Interest on Dutch Water Authority Bond from 1648
Quoth Yale's website:

"Yale does not require students to take out loans for their education. Instead, Yale meets 100% of demonstrated need for all admitted students with a financial aid package consisting of need-based scholarships, term-time employment, and a student income contribution."

http://admissions.yale.edu/financial-aid-prospective-student...

So...Yale does what Stanford does already...

emmett··on Why It's Safe for Founders to Be Nice
Paul would strongly disagree with the idea that a startup is defined by a random list of traits and that you "know it when you see it". He very clearly defines a startup as a company aiming for (and potentially capable of achieving) a very high rate of growth.

Team composition is literally irrelevant.

emmett··on Why It's Safe for Founders to Be Nice
A startup, by Paul Graham's definition, is any business designed to scale up very quickly. A business aiming for say, 5% weekly growth or higher.

http://www.paulgraham.com/growth.html

emmett··on Dinner and Deception – Serving meals to the super-rich left me feeling empty
It's interesting how much it really is a problem of alienation, not of the work itself.

Stage magicians who obsessively practice the same motion over and over own their work, own their output, and feel empowered by the hard work and the success it brings.

Waiters who do the same thing because a larger machine they do not identify with dictates it feel disconnected.

The work itself is a distraction from the real problem, which is that the employee does not believe that what they are doing is important or valuable and that they don't feel ownership over their own actions.

emmett··on Google loses data as lightning strikes
Many people have far more than 10k objects though. Based on that math, if you stored 10 billion objects in S3 (certainly well within the realm of possibility), you'd lose an object on average every 10 years, which is a length of time one can think about.

Your main point (that making high reliability systems more conventionally reliable is building your high wall yet higher still) is definitely valid. But the lossage rate is actually a meaningful number given the extremely large number of objects stored in S3.

emmett··on What’s the best way to spend $20K to help the common good?
Based on this comment I suspect you didn't actually read the article, which explicitly evaluated this option...
emmett··on Ask HN: How to handle tech interviews?
(Caveat: I think the existing interview process is fundamentally broken, but because asking people to code on a whiteboard is suboptimal not because the specific questions are wrong)

I've asked this specific question in interviews before, many times.

It's like a slightly harder version of FizzBuzz. It appears to have near perfect discrimination between "knows enough programming to use nested for loops with if-statements" and "does not". Programmers (good or bad), once the given the formal problem statement, solve it in under 20 minutes. People who don't actually really know how to program will never finish.

You might be imagining a slightly more difficult question, but it really is easy. It's an arbitrary coding challenge, sure, but one designed to not be tricky on purpose. The naive implementation works great.

emmett··on Harj Taggar Is Building a New Technical Hiring Pipeline with TripleByte
Have you ever been involved on the other side of hiring programmers? If so, did you hire them exclusively from their resumes or did you have them demonstrate their ability directly to you in some way?
emmett··on Video games beat interviews to recruit the very best
Please also consider pushing Twitch "Let's Plays" as well. We have a lot of programmers watching!

I agree with what you're saying about video advertising CPMs, though it is quite possible to make a living on Twitch once you add in subscriptions at $5/mo.

emmett··on Ask HN: What up with these startup salaries?
Where are you seeing these numbers from? What stage of startup? If Uber offered you 1% equity and $60k/year you'd have to be an idiot to turn them down at this stage. If an unfunded just-an-idea startup did, you'd never take it.

Basically: you can't evaluate equity in a vacuum. It's entirely company-and-stage-of-growth-dependent. Those offers might be awesome or terrible.

emmett··on $2.25 Lyft line rides in SF
It would be simple for the public to provide this support for seniors and people with disabilities on ride sharing services. Just create a monthly subsidy that can be used for the purpose of any public transit. It doesn't have to be used only on Muni. Problem solved!

In fact, that seems like a great policy idea.

emmett··on Bitcoin revealed: a Ponzi scheme for redistributing wealth
You don't need to spend a bunch of capital on mining equipment to speculate. You can just buy Bitcoin. Usually you buy mining equipment because you think you can make BC cheaper than everyone else with operational efficiency.
emmett··on Don't Talk to Corp Dev
Secondary stock sales means the founder sells a portion of their stock in the company in exchange for cash today.

There used to be a widespread belief in the VC world that you wanted to prevent founders from seeing any cash until the ultimate exit. As negotiating leverage has shifted towards founders this has gone away.

It's stupid anyway, since allowing founders to see a little cash up front actually aligns incentives much better once the company is somewhat successful.

emmett··on Don't Talk to Corp Dev
You don't need to take the meeting. If they're serious, you will get another phone call with "How does $1 billion sound?"

If a company really wants to buy you, saying no will not make them go away. It will make them up their price.

emmett··on Don't Talk to Corp Dev
This is completely, 100%, wrong about how YC thinks about things.

If you've not advised a large number of fledgling startups, you wouldn't know, but really truly talking to corpdev is demoralizing and dangerous. If they really want you they'll reach out with an dollar figure (albiet probably a low one). If they haven't done that, they aren't serious. Don't waste your time.

If actually you've raised from a VC, a small acquisition means you'll not clear preferences, which means you get nothing other than retention. The VC/founder conflict only comes up when you have a medium successful company that could be really successful. But in that case you can almost certainly sell secondary stock, lock in your $1 million today, and still go for it.

But more than that, YC (and Paul) truly want to support the founders whatever they want. We celebrate the small sales as well as the big ones. It's simply not true this is selfish advise that you should disregard.

emmett··on Lecture 16: How to Run a User Interview
Focus groups are about getting feedback on a draft of a product. They're completely different from what I'm talking about, where you're getting feedback on your thoughts on what problems exist.
emmett··on Lecture 16: How to Run a User Interview
It's subtle, but there's a difference between abandoning something because people don't have the problem you're trying to solve and abandoning something because people don't think your solution is good.

Take your Uber example. If you'd done user interviews pre-Uber, it's uncertain whether anyone would have said "yes I want to call a black car on my phone" or not. But I can pretty much guarantee that people would have said "yes, sometimes I have trouble getting a taxi to come pick me up".

User interviews at this stage are about validating the problem, not the solution.

emmett··on Busy Parents Uber Kids Around for a Carpool Alternative
Would you let your child interact with any stranger alone under any circumstance? If so, I don't see what the big deal about Uber or Lyft is in particular. If not, that seems awfully sheltered. We're talking about teens here, not 5 year olds.
emmett··on Our Journey to and Through YC
I think the point of the article was the opposite. The first two times they applied, they were focused on YC. The third time (the time they got in), he says:

"Nothing else mattered, including YC. Application time came around and we thought about not applying. It seems a bit silly to say now (to us), but it is true."

They'd become focused on the business and product, and that was the key.

emmett··on The importance of honoring pro-rata agreements
Hey, I think you might be slightly confused on some terminology.

Pro-rata is about investing MORE money into the company when you raise a round, to avoid being diluted. It's not something you can get "above".

I think you're talking about liquidity preferences here, which are a separate issue. You could get returns "above and beyond" liquidity preferences, so I'm guessing that's what your investors are actually talking about.

emmett··on Startup CEOs who gave up fortunes to turn employees into millionaires
That's a gross exaggeration. If you own 0.1% of the company, you're likely to be diluted to around 0.02% at the very worst, and most likely 0.04% or so assuming 3 rounds of funding.

That's assuming a positive outcome that clears preferences of course - if the company goes in a fire sale you're not going to make anything at all. But assuming a billion dollar exit means you're assuming not a fire sale.

emmett··on Ask HN: Did anyone regret the move from engineer to manager?
Speaking as an engineer who made the transition to management in a big way (from primary web engineer and DBA to CEO with no time to code at all), I have mixed feelings.

I love programming. It's a zen activity for me, turn me loose on a problem and I'll literally lose track of time because I'm so absorbed in the problems. It's one of the purest, most joyous singular activities I've ever had the pleasure of practicing.

After 5 years of doing it 8-10 hours per day, I was getting stale. I was putting in the hours and I think I was productive, but I wasn't encountering new problems very often anymore.

I moved into management shortly thereafter, and being a CEO has been fascinating in its own right. So far (3 years in) I have feel like I'm nowhere near the top of the learning curve. It's a constant challenge, in a way that programming wasn't for me 3 years in. I don't think I've ever been more fully intellectually engaged.

That said, I miss programming. A lot. I find excuses to pick up bug fixes. I wrote an internal tool that manages the distribution of status updates, and at least half the reason was just to be able to code something. I don't regret the career change, but I hope some day I can make programming my main gig again.

As an alternative, I think I could have switched from web programming & DBA work to a new problem domain like graphics or even mobile apps and probably gotten a new lease on learning-curve.

Taking that step back, when I look at where other engineers have gone in their careers within the Twitch organization, the success of a switch into management seems highly idiosyncratic. Some engineers really take to it and enjoy the new challenges; some of them hate it and want to switch back. We've had a number of people try it, decide it wasn't for them, and move back onto the IC (individual contributor) track. We pay top ICs about the same as top managers, so I suppose that might make it easier.

My bottom line advice would be: you don't really know if you'll like management unless you try it. If you're in a workplace where you have the option of switching into a management position, give it a shot. Just know that it's not for everyone.

emmett··on Heroku for Science
Generally speaking, running double the number of samples requires approximately double the amount of work, maybe a little less since you're doing it in bulk.

Are you really saying that going from 10 samples to 11 samples causes a doubling (or 1.2x-ing or tripling or whatever) of the time/work required? That's what exponential means.

← PreviousPage 3 of 30Next →