HN user

mayava

622 karma

I post AI research papers I find interesting

Posts166
Comments18
View on HN
www.topbots.com 7y ago

What are the major reinforcement learning achievements and papers?

mayava
3pts0
www.topbots.com 8y ago

Demystifying ICOs: The Good, the Bad, the Ugly

mayava
2pts0
www.topbots.com 8y ago

Neural Network Generates Fake Startups. Hilarity Ensues

mayava
2pts0
github.com 8y ago

Neural Turing Machines (NTM): PyTorch Implementation

mayava
2pts0
www.msn.com 8y ago

Amazon is deciding if it will make a big move into selling drugs online

mayava
3pts0
sites.google.com 8y ago

ChineseFoodNet: Large-scale dataset for Chinese food recognition

mayava
2pts2
medium.com 8y ago

Introducing TorchMoji, a PyTorch Implementation of DeepMoji

mayava
1pts0
medium.com 8y ago

How to Use Tensorflow and Docker to Create a Production-Ready AI Product

mayava
7pts0
www.botfuel.io 8y ago

Benchmarking Natural Language Processing (NLP) Providers

mayava
3pts0
www.youtube.com 8y ago

The Future of Deep Learning Research

mayava
2pts0
arxiv.org 8y ago

DeepLesion: Automated Deep Mining, Categorization, Detection of Radiology Images

mayava
3pts0
opensource.googleblog.com 8y ago

Facets: An Open Source Visualization Tool for Machine Learning Training Data

mayava
1pts0
www.fastcompany.com 8y ago

There are now fewer black women in tech than there were 10 years ago

mayava
7pts0
medium.com 8y ago

Why Continuous Learning Is the Key Towards Machine Intelligence

mayava
2pts0
www.geek.com 8y ago

IBM Fits AI Wave Forecaster on Raspberry Pi

mayava
4pts0
medium.com 8y ago

What I Learned from Reading Every Amazon Shareholders Letter

mayava
18pts0
www.deepideas.net 8y ago

Gödel’s Incompleteness Theorem and Its Implications for Artificial Intelligence

mayava
2pts0
www.inference.vc 8y ago

GANs are broken in more than one way: the numerics of GANs

mayava
3pts0
www.bbc.com 8y ago

Reality check: is automation worse for men or women?

mayava
2pts0
www.bbc.com 8y ago

Mattel Thinks Again About an AI Babysitter

mayava
2pts0
qz.com 8y ago

McKinsey used machine learning to discover the best way to teach science

mayava
2pts0
medium.com 8y ago

Why We Find and Expose Bad Science

mayava
1pts1
arxiv.org 8y ago

Neural Color Transfer Between Images

mayava
43pts6
blog.piekniewski.info 8y ago

Can Deep Learning Recognize a Cat? The Results Are Not So Obvious

mayava
4pts0
arxiv.org 8y ago

Interpretable Convolutional Neural Networks

mayava
8pts0
www.cnbc.com 9y ago

Apple is working on turning your iPhone into a one-stop shop for medical info

mayava
1pts0
www.technologyreview.com 9y ago

How AI Can Keep Accelerating After Moore's Law

mayava
1pts0
sebastianruder.com 9y ago

An Overview of Multi-Task Learning in Deep Neural Networks

mayava
3pts0
lukeoakdenrayner.wordpress.com 9y ago

The Bleeding Edge of Medical AI Research

mayava
4pts0
www.topbots.com 9y ago

Nuances of Applying Deep Learning to Real World Problems

mayava
2pts0

Hey guys. I'm nearly done hacking up a simple tool that lets me easily stay on top of the social feeds of people I care about without having to filter through everyone else's updates. Basically, I've got aggregated Facebook, Twitter, and Tumblr feeds for 3 close friends sent to my inbox every week. Curious to know if this would be useful for anyone else. If there's interest, I'll iron out the minor bugs, make it look pretty, and release it as a public product.

Pitching is not the be-all-end-all, and you'll certainly get to a point where it stops being so useful for you (i.e. if you've already determined the reasons why certain people aren't interested in your product, no need to pitch them). We're not advocating pitching for the sake of pitching, we're advocating being methodical about and maximizing your learning from your pitches.

For example, biases will definitely crop up when you pitch people. They will also crop up when you go to try to acquire customers or deal with users, so it's good practice to field them early on. Going outside your normal circles will expose you to people who aren't your friends just trying to be nice or "experts" who think they know it all and that your product sucks, so that you can 1) sharpen your own communication skills (you'd be surprised how many startups suck at describing what they do), and 2) refine your understanding of customer concerns.

Also, pitching is just the beginning of validating your idea. For example, if you get people who sound excited and say they'll pay for your product, and then they don't, there can be a number of causes: 1) Maybe they were just being nice, 2) Maybe your product doesn't meet their expectations, 3) Maybe they changed their minds, etc. If you don't pitch these people, you don't have the opportunity to develop the customer insights you can get once they take a harder look at your product.

In the event that someone listens to your pitch and is totally disinterested, this can be valuable too. It may be that the person you thought was a potential customer actually isn't - I'd dig deeper to find out why. Maybe you just need to explain it a different way, or catch them at a better time.

You can get decent data for about $50, but it may take you a few batches to get your questioning and worker filters on just right. I recommend starting with small batches of 20 or less responses, just to test quality (international vs. US only, specific panels, etc)

I highly recommend that for your first set of questions you either use filter surveys (ask for age, location, gender, etc) to target people who are in your target user demographic, or have test questions to make sure they really understood your pitch (ex: Summarize this product in 30 words or less).

In my studies, I limited the respondents to the US, because I was getting too much noise from international workers. I also gave them the option of leaving more detailed feedback. Most of them did, to my surprise.

Twidium is a bit of a skeevy looking tool, I must admit. So far, we've only used it to follow people who have specifically put certain keywords in their tweets or bios, and it has worked well in helping us finding highly engaged consumers. Have you tried using it?

A few ideas: show your product to a diverse base of customers to avoid any super-techy or early adopter biases (unless, of course, your product is mostly for them), quantify the findings, and be systematic about improvements. If there are issues in areas outside your expertise, such as marketing or UX, find professionals you respect who can advise you. Also useful to set metrics beforehand for what you consider success/failure, i.e. # of customers acquired within a certain timeframe, customer growth rate, engagement, etc.

Very true, but there's a difference between showing your customers a product and not listening to them. I agree that the first step is pretty scary, esp if you're an introverted engineer who'd rather hide in a cave and code, but the main problem seems to be that when founders do solicit feedback, they'll hear what they want to hear, i.e. focus on the positive feedback and underweigh the negative. This supports a continued delusion that their product is as loved by others as it is by them.

The entrepreneurs I talked to who were successful in a single founder role were either building low-tech consumer-facing web businesses, or had awesome tech skills and were able to bring on supplementary help at lower equity divisions to be treated like co-founders, but without the hefty equity split. After all, these extra team members were brought on after the companies typically got a bit of buzz and traction, so they had reduced risk.

I wonder what the optimal similarity/difference threshold is for successful co-founder relationships. You'll want to be different enough so that you can merge your expertise and cover more ground, but then if education / background / philosophies / life stage are too divergent, as in the case with you being in uni and your co-founder running a family, you're likely to disagree on major issues or have different levels of commitment/skill.

Thanks for the comments, guys. I'm definitely interested in exploring co-founder relationships further. Seems to me that bringing on a "good friend" as a co-founder is a treacherous path (either take off or crash and burn), but most husband/wife, brother/sister, and very very best friends co-founder relationships seem to work out. Perhaps the extreme closeness allows the parties to communicate more openly, be more committed, or better understand and accommodate each others' weaknesses.