Hacker News Reader

abhshkdzon Scouts by Yutori is now generally available

> Tried it for a real estate search, and to my surprise - it’s better than the “alerts” at major real estate marketplaces. It was particularly useful to me since I don’t speak the language in the area I’m trying to buy.

That's awesome. Glad to hear!

> Just wondering what other use cases there are…

Check out the examples below the fold here: https://yutori.com

abhshkdzon The Age of Agent Experience

Good read, thanks for sharing! I'd love for OAuth to be augmented with agent-friendly scopes. Completely agree that it's a standard that doesn't need to be reinvented. But in how things are today, there's two broad areas where OAuth doesn't quite cut it:

1) long tail of websites that don't have APIs, so the only way for an agent to interact with them on the user's behalf is to log in more conventionally, and

2) even if a website has APIs, there may be tasks to be done that are outside the scope of the provided APIs.

Thoughts?

abhshkdzon Using AI to discover catalysts for use in renewable energy storage

Whitepaper: https://dl.fbaipublicfiles.com/opencatalystproject/papers/oc...

Dataset: https://dl.fbaipublicfiles.com/opencatalystproject/papers/oc...

abhshkdzon Google Buys Fitbit for $2.1B

Garmin

abhshkdzon Launch HN: Hatchways (YC S19) – Internships Instead of Interviews

This seems to have tremendous potential. I like the conscious effort to de-bias the system away from initial indicators (like school, GPA, etc.). Thank you for doing this!

Question -- is there the possibility of an income-share model here? i.e. Hatchways does due diligence in initial assessment + matching and even pays for the candidate's internship at the company, and makes money only when the candidate gets a full-time offer as a pre-agreed upon percentage of the candidate's salary (for the first N months say).

abhshkdzon “Only five people maintain Python-pip”

Well, thank you pip people! <3

abhshkdzon Twenty-one cities in India will run out of groundwater by 2020

More context on this: https://twitter.com/jslaternyc/status/1144483586042548225

abhshkdzon Habitat: A fast 3D simulator for embodied AI research

From what I understand, this is not a physics engine (at least in its current form). This is a bare metal 3D environment rendering engine particularly optimized for speed; speed being especially important for training deep reinforcement learning agents for instance.

Physics would be a layer above this; I imagine Bullet can be integrated within Habitat, but will come with significant hit on speed. These choices depend quite a bit on the task being studied.

abhshkdzon Habitat: A fast 3D simulator for embodied AI research

Not quite. Real-world environments (which Habitat is aiming to be a close-enough proxy for along the computer vision axis) are far more visually complex than what Doom allows for.

abhshkdzon Habitat: A fast 3D simulator for embodied AI research

https://aihabitat.org/

https://github.com/facebookresearch/habitat-sim

https://github.com/facebookresearch/habitat-api

abhshkdzon List of Machine Learning/Deep Learning Conferences in 2019

Plug: https://aideadlin.es for academic conferences

abhshkdzon Ask HN: What are your predictions for the next 10 years for our daily lives?

- AR: Handheld devices will make way for wearables with augmented reality displays. This will largely be led by usability advantages. Like Google Maps will see a shift towards glasses with augmented reality displays. Think navigation instructions from Google Maps overlaid on street view-like camera imagery on a wearable device.

- Robots: Robot sidekicks, like Vector: https://www.anki.com/en-us/vector, but bigger and more useful, will become common. Tbh Vector is currently at best a cute and expensive toy. With progress in AI, specifically at the intersection of language understanding, computer vision, and robotics, these robots will become easier to talk to and work with.

- Self-driving cars: Self-driving cars will be common, with some business models getting to wider deployment faster than others, for e.g. Waymo, Uber, Lyft. New set of traffic protocols will likely emerge on 100% self-driven localities (humans' traffic rules are _extremely_ inefficient for robots) for faster + safer transport.

- Space: Space tourism will start to emerge. There will emerge entire industries supporting this shift -- from ticketing aggregation services (like Kayak.com, etc. for space) to data services (equivalents of AT&T, TMobile, etc. for space) to cargo (faster earth-to-earth, earth-to-x), hotels, hospitals, etc.

abhshkdzon Volta: Advanced Data Center GPU

I think Jensen mentioned this would be available with OEMs Q4 onwards

abhshkdzon Volta: Advanced Data Center GPU

DGX-1 with Volta — $149k, Q3; DGX Home Station with Volta — $69k, Q3

abhshkdzon Visual Dialog

And here's a video overview: https://vimeo.com/193092429

abhshkdzon StackExchange Tagger

I see, okay. Thanks!

abhshkdzon StackExchange Tagger

Great work. It might be a good idea to add usage instructions to the GitHub readme. A few questions:

— Are you releasing the dataset? Is there any place I can get more details about the dataset? I'd love to see how well a simple RNNLM does.

— Is it an end-to-end pipeline? As in, if I have a SE question url or text, can I just pass that in and it'll automatically do all the preprocessing and give me the predicted tag?

— Is the trained model available somewhere?

abhshkdzon Visual question answering using CNN+RNN

As it stands currently, it's quite far off from cracking captchas. :-)

abhshkdzon Visual question answering using CNN+RNN

Accuracy is measured as min((number of humans that provided that answer)/3, 1) i.e. 100% accurate if at least 3 humans provided that exact answer, as outlined here: http://visualqa.org/evaluation.html.

No, this model is from the NIPS15 paper by Ren et al (http://arxiv.org/abs/1505.02074).

abhshkdzon Visual question answering using CNN+RNN

Priors from the language play a much bigger role in the answers that are predicted than the image itself. So for example, if you ask 'What color is ...?', irrespective of the image, it is more likely to spit out colors as the answer. The answers are usually well-aligned with the question that is being asked. 'Yes/no' for binary questions, 'red/blue/etc' for 'What color...', 'tennis/baseball/etc' for 'What sport...' and so on.

abhshkdzon Visual question answering using CNN+RNN

Although there is no catch, it's far from perfect and hardly magical. Its accuracy goes up to ~55% on the VQA (http://visualqa.org/) dataset (which is short of state-of-the-art by ~7%).

abhshkdzon Google Open-Sourcing TensorFlow Shows AI's Future Is Data, Not Code

Aren't we getting too ahead of ourselves here? It's just been released, we don't even have favorable benchmarks yet. Sure, it's by Google and lots of big names are associated with its development, but it still has to be adopted by the community and proved to be faster/easier/more flexible than Torch/Theano.

abhshkdzon Erdős: Mathematical problems

Bingo, there was some discrepancy, so we had to take #43 off.

abhshkdzon Erdős: Mathematical problems

Not necessarily, like you rightly pointed out. Some require computation. while others can be worked out by hand.

abhshkdzon Erdős: Mathematical problems

Hey, thanks for the feedback!

* About the app: http://erdos.sdslabs.co/about * Added 'Activity' to topbar.

Subscription to new problems is a cool idea. I'll add it to my to-do list.

abhshkdzon Erdős: Mathematical problems

Yes, that is exactly what it is. I'll create an 'about' section soon. Thanks :)

EDIT: http://erdos.sdslabs.co/about

abhshkdzon Erdős: Mathematical problems