create a persona that can have a conversation with people

see the movie "Her"

http://www.imdb.com/title/tt1798709/

how many people will fall in love with Watson?

Let's call it Emma Watson then.

I already do, and don't think I'm alone...

your idea might be better. I could help if you need support.

your contact? check my profile for twit id

hi, you can email me at: timothee . henry @ gmail.com

One idea: it might be worth asking for email addresses. You already had 27 people interested. Build a quick unbounce page and get the emails so you can recontact.

Excellent idea.

Not completely sure what scope you have in mind.

It reminds me of two things:

1. Tools of the Trade: https://news.ycombinator.com/item?id=5235137

they do an edition per year.

2. and somewhere there is a google spreadhsheet of all hacker news derived websites (all the guys who scrape hacker news to build "derivatives").

Thanks for the link. Great to seed the initial db.

http://hn.algolia.com/api Thanks. How many requests would that require ?

http://algorithmia.com/demo/hn does not work for me.

The first link states at the bottom: "We are limiting the number of API requests from a single IP to 10,000 per hour. "

The hn/tag demo works one time out of five refreshes for me, so keep trying. Here's a screenshot in case that doesn't work: http://imgur.com/yPF0hkn

Thank you. I appreciate the screenshot. Interesting; I get the feeling the tag algorithm is crucial to make it work. Not sure it's such a clear no-brainer as I first imagined it...

The contest actually puts the burden of proof on the "claim of insecurity" side.

The contest is a stunt that appears to a casual observer to shift the burden of proof because they wrote a long winded "PROVE ME WRONG BRO."

> They use the broken SHA1 hash function

They answer this point on the website: "Q: Why do you use SHA-1 in the place of a MAC? [...] since this means still requiring at least 2^128 operations (instead of 2^256 with, say, SHA-2) to even begin trying to break this scheme, the trade-off seems fair."

Why not break the crypto (and take the money) if it's so amateurish?

The easiest to understand response to this question that I've seen so far is from this comment [0]:

The contest limitations rule out most of the likely attack vectors for breaking the protocol in the real world. It's like saying "Our bank vans are 100% secure. Just try stealing money from them without puncturing our tires or bribing one of our employees."

[0] - https://news.ycombinator.com/item?id=6936949

In particular none of the attacks described in TFA (Known Plaintext, Chosen Plaintext and Chosen Ciphertext) are possible within the frame of their contest (since Telegram controls all inputs).

Yesterday someone blogged an example of a completely broken cryptosystem that would still pass Telegram's challenge with the same limitations: http://www.thoughtcrime.org/blog/telegram-crypto-challenge/

That's Moxie Marlinspike, developer of TextSecure.

With a very valid challenge.

It may even be possible to factor the RSA Key

More to the point, KPA,CPA, etc are very important, and systems should be definitely tested against them, but in real attacks, they may not be available

""It's legalized front-running. I think it is basically evil and I don't think it should have ever been allowed to reach the size that it did," he said. "Why should all of us pay a little group of people to engage in legalized front-running of our orders?""

Charlie Munger, http://www.cnbc.com/id/100705820

What's the difference between HFT and traditional Human Market Makers?

May I suggest that Electronic trading and HFT have made it possible for a few people to do the work of thousands. This has brought down the cost of trading.

Charlie's just upset that some of his bridge partners aren't scalping what they used to with their floor-based market making businesses.

Which HFT firms have access to the orders of other participants before the exchange does?

I'd like to know more about slide 6:

http://amueller.github.io/sklearn_tutorial/#/6

Why the [Classification][100K sample?] checkpoint?

And more info in general about this whole cheat-sheet.

SGD is fast but not necessarily more accurate. If you've got lots and lots of data, then a simple yet fast approach is likely to be a good choice to start with.

Thanks.

I also found a blog post that figured the diagram, with some background infos:

http://peekaboo-vision.blogspot.de/2013/01/machine-learning-...

And a prior HN discussion: https://news.ycombinator.com/item?id=5831512

There are many others too, but with few points and comments

A result example would be nice.