teh

I read somewhere that a single machine still wins 25% against alpha Go. That's still 500 W VS 20 W but a bit closer.

A friend of mine works as a beetle curator. Telling species apart is actually surprisingly difficult. He shot down a lot of my technical solution suggestions with ease. Genetic sequencing might work some day but it's not there yet (and they have a lot of specimens, like 10-20 million).

E.g. there is a species where each specimen looks very different (e.g. size between mm and cm) depending on how much food they consume. The best way to differentiate is often to extract the genitalia.

One thing the article doesn't explicitly mention but alludes to is that a lot of species we think we know may actually be different species, or just the same species with a different look. A lot of the descriptions are 50 years old and read "black body with six legs and a head". So all those need to be re-done at some point.

For us the free doesn't matter because certificates are cheap. It's the byzantine and insecure process of obtaining a cert (sending us our private cert in a zip in a plain text mail. I mean, really?) that makes LE so great.

Edit: Total brainfart, apologies; The company sending us private info as a zip was a different thing.

Do you have pricing information available anywhere? All I can see is [1] but that's not really interesting. Compare with [2] which makes it quite easy to guess how much I'd pay.

[1] http://www.alchemyapi.com/products/contact-sales

[2] https://cloud.google.com/vision/pricing

We're using Nix in production. It has some polish issues but even in its unpolished state Nix is adding a lot of value for us.

I meet people that have all kinds of familiar issues with puppet, chef etc. We used to have those problems and we just don't have any more.

Shuffling can be done in linear time, e.g. with https://en.wikipedia.org/wiki/Fisher%E2%80%93Yates_shuffle so yes.

This is not what's happening though AFAIK. In the lottery illustration there will be a single payout. In derivatives the risk of default for a certain "class" is treated as IID, and banks can leverage a certain amount on different risk classes (100x is not unusual).

But as 2007 demonstrated the real world risk is not IID. And at 100x leverage a 1% movement can totally wipe out a bank.

Also it's not like the 2007 crisis was fixed in any meaningful way. The FED just bought 1.4 trillion of MBS derivatives and is now sitting on them, actual value still unclear. Probably to be paid by the tax payer through 20 years of austerity.

Super interested! I often find technical blog posts like this years later and I'm really grateful to the people who donate their time to the difficult task of writing them.

Transferwise already require receipts for some payments. I recently ordered something from Asia where payment went through an intermediate European payment provider and Transferwise wanted a receipt from me to proceed.

Python's print method is actually fairly clever picking the "nicest" representation that's still correct.

https://docs.python.org/2/tutorial/floatingpoint.html#tut-fp...

Quote:

"In current versions, Python displays a value based on the shortest decimal fraction that rounds correctly back to the true binary value, resulting simply in ‘0.1’."

Can you elaborate which ones you're missing?

The R&D costs argument is a diversion. Profit margins for Pharma companies are consistently 3-5x of more competitive industries (e.g. car makers [1], [2]). Health is a failed market full of rentier corporations.

It's hard to argue against them because they are doing good (inventing life saving medicines) but the opportunity cost of money paid out as profits instead of reinvested into research must be huge.

[1] https://ycharts.com/companies/VLKPY/gross_profit_margin https://ycharts.com/companies/BAMXY/gross_profit_margin

[2] https://ycharts.com/companies/PFE/gross_profit_margin

I'm one of the people running passopolis. We think that Mitro was already pretty good but we've fixed several bugs, packaged it for the Chrome store, made sure the server runs reliably etc. As the FAQ explains we changed the name to avoid confusion when Googling.

We're also not building a company around it, we've absorbed the work of keeping it running as our agency (wearewizards.io).

If we start charging it's going to be for some actually new feature, not for the current product.

I'm running a small agency with two friends and we're keeping mitro alive (and better: are fixing issues) here: https://passopolis.com/

Firefox and Chrome extensions are working, and I'm currently spending a few hours a week on migrating to the new Firefox-extension protocol.

It's a super interesting topic to explore. Investigating how much bang we get for a buck in medicine is a touchy subject because when I'm ill I want the very best but while I'm fine I'm more willing to consider trade-offs.

Keeping that in mind I think that averaging numbers is not a great representation of doctors impact, the same way average latency isn't a great representation of what users see. What's the 95% percentile of QALYs? E.g. are doctors making one life (e.g. kid with a broken leg as mentioned in a comment here) better for 50 years when 9 people will just never break a leg?

If looking at shifts in distributions the story might (I could not find raw data) look very different.

Hey, I just signed up and I'm a bit unclear on how the "reproducible" part works. If I download my files they don't contain e.g. a requirements.txt?

My gut reaction is that I'll be paying money to install my telescreen?

I have this slightly panicky feeling that my perception of what's acceptable sharing of private data is so disconnected from current popular sentiment that I'm becoming a bitter outsider.

I'm sure it's technically well executed as expected from Google management but I do hope it bombs as a product for (to me) obvious reasons.

Can I ask which library you are using? I found a few [1] but none seem super robust.

Also, how do you deal with the bytes type?

[1] https://code.google.com/p/protobuf-json/ https://github.com/benhodgson/protobuf-to-dict

Not sure transit is designed for the same space. E.g. there seems to be no schema, and the default JSON encoding isn't super readable either.

Protobufs can be encoded as JSON and as text, so there are some ways to address the readability I guess.

Another issue is full paths to the nix config file encoded in the state file itself. The state file should not point to the config files used IMO.

I'm not sure I understand how image grammars (whatever that is exactly) suddenly pop up as a solution after such a long introduction. I could not find any evidence or literature about them in relation to learning.

The author is stating a hypothesis but no way to test it. I'm not sure what point he's trying to make.

Today I looked at Nim in a bit more depth because it keeps popping up. I have a slightly uncomfortable feeling about it that I hope is unfounded!

To me it looks like it makes the unsafety of c more accessible because of better tooling and nicer syntax. Looking at e.g. [1] there are still pointers, null-pointers etc, just like in c. So now you have a language that looks superficially simple but is actually very dangerous. Compare this to e.g. rust which was the most painful thing I learned recently but I also know that it brings something fundamentally new to the table.

Anyway, there's a lot I don't understand about Nim and I'd be happy to see evidence to the contrary.

[1] http://nim-lang.org/0.11.0/tut1.html#advanced-types-referenc...

I thought it was trivial to quickly compute SHAs so you can try lots of passwords.

In the case of sending bcrypt output wouldn't the "password" be long and essentially random and therefore not easy to enumerate?

Or is SHA also reversible?

Hypothesis looks fantastic. A friend of mine and me spent a while on Google collecting Python QC implementations but didn't find any that implemented shrinking.

Hypothesis escaped our search. I don't know how but the project could do with better Google ranking :)

It has some really useful features like storing the counterexamples.

The CLA is a bit odd but I'm sure you have good reasons!

IANAL so I didn't get that. Thanks for explaining!

What makes the patent grant unfair? I think saying "you can use our pretty powerful technology until you sue us" is a great way to defang the patent system.

[1] https://github.com/facebook/react/blob/master/PATENTS

Does anyone know what this means for spot instances?

Interesting! When I check the DOM in inspector I can see it refreshing - maybe that's a bug?

It's nice to see Microsoft cares. OTOH part of the attraction of the RPi for me is that people freely share their knowledge, code and cool hacks.

Adding a humongous blob of unmodifiable software doesn't really align with any of that. In the best case people will shun this, in the worst case it could harm the culture.

The learning curve is pretty steep. It's not just a new config language, it's also letting go of many preconceptions one may have (where stuff lives on the filesystem e.g.).

That's also true for software I'm packaging. Certain ecosystems just really love to load configuration files and write log files relative to the binary path which breaks in nixos because all packages live on a read-only mount (looking at you ruby, cargo from rust will also be fun).

Lastly there's just not the same number of contributors yet so you're going to have more of the small issues that get fixed and reported by the armies of users that e.g. Ubuntu has.

I'm saying all this as an incredibly happy user. I can think of few technologies that have given me as much leverage as nixos in recent years.