Follow Slashdot blog updates by subscribing to our blog RSS feed


Forgot your password?

Comment Seastar or C*? (Score 1) 341

The similarity of the name 'Seastar' to Connection Machines' dataparallel programming language C* can't be an accident. But C* needed to run in shared memory or at least atomic synchrony on low latency distributed memory in order to preserve consistency. And of course, it needed SIMD algorithms (do the same op concurrently on a large pool of data) or it could add no value over using C.

Sounds like a misnomer to me.

Comment Re:Because (Score 1) 65

I hate to admit what you say seems to be true at all big corporations. At the giant pharma where I work I've seen less and less S/W innovation take place internally in the past decade. This has had two big side effects: 1) all our best computing have left, and 2) so has all the interesting work.

There's no longer any interest or even tolerance among managment for novelty or invention in-house (AKA risk). Skill development focuses on the project management side only; no tech. All IT has to be done externally, from data mining to software dev to app integration to sys admin. Nothing is done in-house anymore except invitation of external S/W vendors and external integrators and support, then monitoring their progress until the system is installed. A year-end accomplishment is check-boxed and all worker bees hum with one voice, "Booyah".

Ten years ago computing was different. May IT RIP.

Comment OK, it makes sense now... (Score 2) 403

...look at the source. The caption of the article's main photo tells it all:

"The directors of the FBI, CIA, NSA, NGO, DIA, and NRO stand for a group picture with Fox News' Catherine Herridge (second from left) and executives of INRA and AFCEA at the conclusion of their panel discussion at the Intelligence & National Security Summit in Washington on September 10."

The supersilly quotes were directed at Fox News viewers. They were never intended to be taken seriously.

Comment Self promotion. Move on. (Score 4, Informative) 58

1) A camera programmed to identify objects then speak the label aloud is NOT sentient. It isn't even AI. It's computer vision technology from around 1995. An Amazon Fire phone can do far better and nobody claimed it was sentient either.

2) This pair are terminal Master's students in "Professional Studies" and "Software Engineering", not "AI researchers". Clearly their future lies in advertising and politics, not AI.

Blame the Motherboard author. Nothing to report here. Move on.

Comment Re:Actually no (Score 2) 165

I know more about nonhuman studies than clinical, but according to the US HHS (who runs FDA), the breakdown of costs are these:

- $15k/patient for phase I
- $20k/patient for phase 2
- $25k/patient for phases 3 and 4

The cost of the average trial:

- phase 1: $4 million
- phase 2: $13 million
- phase 3: $20 million
- phase 4: $20 million

Some phase 3 trials can be larger and last longer than average, like 20,000 patients over 5 years. Obviously at the average cost of $25k/patient, such a trial would cost $500 million, well over the average. In fact, a long study can greatly increase the per patient cost as well.

Because multiple trials are run in each phase for each drug, these trial costs are multiplied.

The principal cost in any trial are the medical procedures (~25%): drug administration, tests (lab, imaging, biopsy, etc), exams, etc. These are repeated multiple times on each patient during each trial to monitor changes in both efficacy and safety.

Here's a thorough accounting from US HHS:

These costs are set by FDA regulatory standards and the medical laws of each country where the trial is performed. Of course if you want approval for your drug in another country, you must comply with all their rules as well, often repeating studies using their residents (e.g. Japan).

This 2012 Forbes article by Avik Roy offers further insight on why clinical trial costs are rising:

Pharmas must play by these rules, but they don't write them. Lawmakers do that.

Comment Re:Pharma development is hard and expensive (Score 3, Informative) 165

I do work in pharma, and your sequence of steps sums up the process nicely.

I'd add that for every drug that succeeds, roughly another 20 fail, often after 5 or 10 years of development and costs incurred. That's why the estimated development cost of each new drug is widely acknowledged to be a minimum of $1 billion US (though most cite $2B as the norm). However after you include the cost of all failed drugs, the cost of producing each drug that succeeds effectively rises to between 4 and 5 billion. This is why each new drug needs to be a big selling blockbuster. It has many mouths to feed.

Obviously open software and volunteerism has their work cut out for them if they are to make drugs affordable. But I *would* be curious to know where their advocates believe these forces could have significant impact. It'd have to be in the clinical trial phase, where 80% of cost is incurred.

(BTW, to compute the net average cost of each new drug, you divide pharma company annual R&D budgets by the number of approved drugs/year. Matthew Herper of Forbes has covered this topic extensively, as has pharma chemist chemist Derek Lowe in his blog "In the Pipeline").

Comment Re:Actually no (Score 2) 165

Yes. Clinical trials are famously expensive -- no less than $100 million US for any drug that is not fast tracked, which reduces development time (and cost) by no more than half.

In general only untreatable mortal diseases like cancer or infection can fast track a drug. The other 95% of drugs go through probably 5 years of compound identification, tuning, and testing, then 5 years of preclinical trials in multiple animal species, then another 5 years in humans before approval. (Yes, that's about 15 years.)

The last phase of development (clinical trials) is unavoidably very expensive (80% of the overall cost). And no amount of free software or volunteerism is going to change that appreciably. Even now, all patient participants in clinical trials are already volunteers.

The cost of a new drug lies in planning, testing, write up, peer review, and great gobs of regulatory oversight and process. Open source cannot change that.

Comment Re:Pharma development is hard and expensive (Score 1) 165

Nonsense. Why should all pharmas conspire to raise the cost of clinical trials? Why would the FDA agree to this? Why would every other country and their version of 'FDA' play along, when a single dissident would cause such a house of cards to crash instantly?

As it happens, I *do* work at a giant US pharma, so I know how unworkable such a scheme would be.

Comment Re: There's an easy solution to this problem...Tru (Score 2) 214

Agreed. The novelty and utility of not having to own a car will more than compensate for the added inconveniences that the fleet owners will require when they first arrive on the market.

Adter all, something like 4 million people now make their living driving cars and trucks and buses. They and their unions will put up a hell of a fight against automation. Fleet owners will have to bend backwards to allay the many threat scenarios proposed. Validation of driver ID and car passenger is a very small bump in the road.

Comment Re:There's an easy solution to this problem...True (Score 1) 214

I think the TSA is an effective counterargument to your overconfidence that people will accept that risk. Requiring the removal of belts, shoes, watches, and anything steel shows the absurd lengths bureaucrats will go to when overreacting to threats, even very rare ones. I'm sure the giant corporations behind AV cars will be comparably risk averse. After all, should someone actually deliver a bomb in such a car, they could see an immediate end to their entire business, or such a severe curtailment, stockholders could lose faith and sell off.

No, the adoption of AV cars will be gradual and become easier as everyone learns their limits. Initially, the rules for their use will be stricter. As the tech and infrastructure improves, their use will broaden and more variatons will be permitted.

For instance, I'm sure children will not be able to ride unattended until the system gets a few million miles under its belt. The same is likely for unattended package delivery. All it takes is one bomb in one tunnel...

Comment Re:There's an easy solution to this problem...True (Score 1) 214

Actually, I think the parent poster is right. AVs can be set up so that the customer can't send the car to a destination. In the early days of AV cars, no package deliveries will be permitted without a person riding in the car who can answer authentication questions en route.

Also, when renting the car, probably you will be required to show a preregistered ID, and perhaps a message will be sent to your cell phone requiring further authentication responses.

No. I think AV cars *can* be made acceptably secure. But it'll be a little tricky and sometimes annoying, requiring some of the conventions we use now when renting a car.

Your good nature will bring you unbounded happiness.