Kary Mullin may be an exception -- he did something (invent PCR) that seems to put him on a pretty high achievement plane (quick Nobel Prize, key technology for all of biotech, etc), and from his own accounts, he didn't really work obsessively long or hard on it. He was just driving home after a long trip, stopped by the side of the road, jotted down the basic idea for PCR, then spent a year of 40-hour weeks in the lab getting the bugs out of it.
The 1000g + 1000ug figure was all the students total at Caltech -- for maybe 200 professors or so, I guess. So, as a fraction of campus, his post-doc crew was about equal to about 7% of the entire grad student population. Since many of the research groups had no postdocs at all (our group had about 1 post-doc at any given time, and 6-8 grad students, and we were considered a bigger group in our department), his group was basically 10X the size of your typical large Caltech research group.
As someone who was around the Caltech campus at the time the article describes, its important to note that Lee Hood was very atypical for a Caltech prof -- his lab was all post-docs and huge (at its peak, around 70 post-docs), this on a campus that has about 1000 undergrads and 1000 grad students, and maybe a few post-docs per group depending on the field. He was a much better fit for a big med school than he was for Caltech. Caltech...
May 1993, SCOC was trading at 12 dollars a share. Today, its under 8 (no splits in the interim). I don't think we can accuse these guys of maximize shareholder value for the long term...
Newspaper lockboxes are often cited as an example of the "people are basically honest" dictum with respect to IP. The common street box in the US for buying newspapers lets you take an arbitrary number for your quarter, with a vanishingly small chance of being apprehended. Yet, losses from people taking extra papers are so low, that there is no economic incentive to replace the boxes with smarter technology. In a dishonest society, the first guy each morning would take all the papers and hand them out for free at work.
So, just how much patenting can Celera really do with this shotgun data? Especially if the public folks are looking at their half of the shotgun data, making inferences about function, and putting it in the public domain in parallel?
The big worry seems to be that they could be a tollkeeper for a large fraction of all genomic research via this patent route -- but its unclear from the press reports just how realistic this fear is...
MSFT has already given up its leadership role as a tech-stock, an adverse DOJ ruling isn't going to cause a 1987-style market crash.
But if John Chambers (Cisco CEO) gave an earnings warning that implied that the growth rate of internetworking hardware was slowing down to the growth rate of Bud Lite sales, put a fork in the longest bull market in history, its done.
MPEG 4 Structured Audio is a good base language for building music apps -- our web page has an online tutorial about the standard, and a MP4-SA to C translator that produces runtimes that can work in real-time. It's a "meta-answer" to the Ask Slashdot question; it's probably not going to help today, but I think its the right techology for Linux to build on.
This topic reminds me of the middle ages, when it was thought commoners just couldn't be expected to learn how to read. So elaborate art was commissioned to teach the Bible via pictures, and the rest was handled through the oral tradition.
Oh, and not teaching the commoners to read had nice side effects for the upper classes too. Then came Gutenberg, and the next thing you know you Tony Blair is tossing out the inherited members of the House of Lords:-).
Fast forward to the 21 century, we see history replaying itself -- now commoners are considered too stupid to program instead of too stupid to read. Having watched secretaries with high-school educations become TeX Macro wizards, I can say the commoners are pretty damned smart, and are starting to notice they have nothing to lose but their chains.
If your application can be fielded entirely in ANSI C and its libraries, you can just use dos-gcc to compile up a DOS executable under Linux (dos-gcc is djgpp built as a cross-compiler). I build each sfront release under Linux for both Linux and DOS/Win32 using dos-gcc, and it works great. It was worth purging gratiutious non-ANSI-libs from the source tree...
Some of us developed Open-Source UNIX console apps that are strict ANSI-C. Without leaving the comfort of our UNIX machines, we can use dos-gcc to make.exe files to distribute along with our tar.gz's and rpm's.
My question to you is, how do you get the rest of the "culture" right for the DOS distro, if you're unwilling to immerse yourself in DOS by using it? I'm basically talking about doing all the stuff that "rpm -i" does automatically on a Linux box, along with issues like portable uses of the ANSI system() and getenv() to do things like find a local C compiler and execute it from your C program. Is there a HOW-TO or FAQ about making your UNIX console program fit (as opposed to work) in a DOS world? Thanks in advance!
P.S. To put the question in perspective, the app I actually maintain is a program that generates C files, thus the compilation issue.
When I was a grad student at Caltech in the 80's, we used MOSIS (www.mosis.edu) as a silicon broker to fab IC chips. It was all done via the Internet, we sent CIF design files via email, and automatic responders sent update messages tracking the chips progress. Three months later, a Fedex box would show up with the test chips, and a goverment grant (and before the 80's ended, a VISA card of a company or individual) got billed. So, I'd say Internet shopping worked quite well pre-PC, as long as you were buying IC fab services:-).
At AES last year I went to a Dobly demo room, where they remixed different types of music for 5:1. Musically, I can't say it added anything to the experience for me.
Now, for movies and video games its different, placement around the space has meaning in the context of the story, but there's a long tradition of musicians all being in one place -- on a stage -- and people sitting in front of them, which is a good match for stereo placement.
Nesheim, John L., 1942- High tech startup : the complete how-to handbook for creating successful new high tech companies / John L. Nesheim. Saratoga, CA : Electronic Trends Publications, c1992.
Found in good university bookstores (Stanford bookstore has it), or at amazon.com
Analog neural nets work best in special purpose application -- where what you want to do is to take a stream of input vectors, feed it into a MLP, and generate output vectors, ideally at the maximum rate that the analog circuits work. Or, if you're into power instead of speed, push down the power used by the net until you're just barely making the timing. This works the best when there's integrated sensors (like photoreceptors or a silicon cochlea) or actuators (micromachined stuff) on the chip, so you dispense with an A/D or a D/A by going analog as well.
There are a few products out there that use analog nets just in this way -- there's one in your Logitech Marble Trackball, computing the motion vectors of a pseudo-random dot pattern on the ball in IR.
But most problems out there actually don't fit in that nice little niche -- instead, the neural net is part of a signal processing chain, and the adjacent steps don't fit well in analog, so you need a DSP anyways. And once you have the DSP, its usually is a lose to put an analog neural net on the die, instead of just adding another ALU to the DSP, since the rest of the system is digital anyways.
One hope is this networking protocol, the address-event representation, that lets analog neural net chips communicate with each other in a very digital way, while being very efficient to implement in analog. See this paper for details...
I think its important to remember our roots here... I know when we put out our E-CAD package under the GPL in 1990
www.cs.berkeley.edu/~lazzaro/chipmunk
we weren't doing it for "world domination", we were doing it so there would be nice (for the time) tools for people to use to design chips. And from what I can tell, a lot of the spirit of gcc, emacs, Linux, ect... is just that -- putting out nice tools by and for a community. If we keep the focus on quality and community, and not the folks at Redmond, only good things can happen...
How do killer apps happen? To use Geoffry Moore's analogy, they are the result of the Tornado -- technology and users needs coming together at just the right point in time, so that a lazy product segment becomes an unstoppable force. And in the end, one company ("the gorilla") dominates the segment. Looking back in history, relational databases were a tornado, Oracle became the gorilla. Internet infrastructure was a tornado, Cisco became the gorilla. Desktop publishing was a tornado, and the Macintosh and Adobe shared joint gorilla status until recently.
On the free software server front, Apache and Samba were the real gorillas, and Linux came along as infrastructure for them. If we're looking for desktop "killer apps", we need to look to the future, predict the most probable Tornados, and be ready -- not look in the rear view mirror at Office, Microsoft got there first. Especially since most of us don't even use an app like Office to get our own work done -- we use TeX+emacs.
> Keeping a company private doesn't > do much to the share price.
In practice, if an IBM carries the stock of a small public company on their books, it's not reflected in IBM's stock price in a material way. And if an IBM sells the stock, its seen as a "one-time gain", unimportant in analyst's eyes. Strategic reasons dominate in investments like this, and an IBM's preference for RedHat going public or staying private lies in strategic issues, not bottom-line issues.
Coca-Cola == RedHat RC Cola == Cheapbytes RedHat distribution
In the end, cola is cola -- why do the vast majority of people walk past the RC Cola and buy the more expensive Coke? Brands are powerful things. RedHat is setting themselves up to be the 21st century Apple -- building a brand that means high-tech rebellion, just like Apple did.
But what does Ben Rosen know about managing a computer company where you sometimes have to design the key components (hardware and software) yourself? It's like taking the CEO of Rite-Aid drug stores and putting him in charge of Merck. And if Rosen doesn't get it, the person he hires to take over probably won't either... just like Pfieffer didn't.
There never really was much synergy with buying Tandem and DEC, it harked back to an era when CEO's didn't get humunous stock options, and so they built empires to gratify their egos. Watch for a trivestiture with the new CEO...
> The problem is that we have presented no > alternative economic model for companies > who truly want the advantages of the bazaar > --but still have bills to pay.
It's imporant to note that several types of occupations, from airline pilot to modern dancer, have become harder to justify doing in purely economic terms, because there are enough people who want to do them who aren't primarily concerned about total monetary reward amortized over the career. As a result, people for who the "bills to pay" issue is significant simply do something else for a living. Maybe a part of what we're seeing here is certain types of computer programming entering this territory...
Slashdot Mirror
Can't believe the Viterbi algorithm didn't make this list ... without it, so much communications and data storage would be slow and/or unreliable.
Kary Mullin may be an exception -- he did something (invent PCR) that seems to put him on a pretty high achievement plane (quick Nobel Prize, key technology for all of biotech, etc), and from his own accounts, he didn't really work obsessively long or hard on it. He was just driving home after a long trip, stopped by the side of the road, jotted down the basic idea for PCR, then spent a year of 40-hour weeks in the lab getting the bugs out of it.
The 1000g + 1000ug figure was all the students total at Caltech -- for maybe 200 professors or so, I guess. So, as a fraction of campus, his post-doc crew was about equal to about 7% of the entire grad student population. Since many of the research groups had no postdocs at all (our group had about 1 post-doc at any given time, and 6-8 grad students, and we were considered a bigger group in our department), his group was basically 10X the size of your typical large Caltech research group.
As someone who was around the Caltech campus at the time the article describes, its important to note that Lee Hood was very atypical for a Caltech prof -- his lab was all post-docs and huge (at its peak, around 70 post-docs), this on a campus that has about 1000 undergrads and 1000 grad students, and maybe a few post-docs per group depending on the field. He was a much better fit for a big med school than he was for Caltech. Caltech ...
May 1993, SCOC was trading at 12 dollars a share. Today, its under 8 (no splits in the interim). I don't think we can accuse these guys of maximize shareholder value for the long term ...
Newspaper lockboxes are often cited as an example of the "people are basically honest" dictum with respect to IP. The common street box in the US for buying newspapers lets you take an arbitrary number for your quarter, with a vanishingly small chance of being apprehended. Yet, losses from people taking extra papers are so low, that there is no economic incentive to replace the boxes with smarter technology. In a dishonest society, the first guy each morning would take all the papers and hand them out for free at work.
So, just how much patenting can Celera really do with this shotgun data? Especially if the public folks are looking at their half of the shotgun data, making inferences about function, and putting it in the public domain in parallel?
The big worry seems to be that they could be a tollkeeper for a large fraction of all genomic research via this patent route -- but its unclear from the press reports just how realistic this fear is ...
MSFT has already given up its leadership role as a tech-stock, an adverse DOJ ruling isn't going to cause a 1987-style market crash.
But if John Chambers (Cisco CEO) gave an earnings warning that implied that the growth rate of internetworking hardware was slowing down to the growth rate of Bud Lite sales, put a fork in the longest bull market in history, its done.
Oops, that's here for the online book on Structured Audio.
MPEG 4 Structured Audio is a good base language for building music apps -- our web page has an online tutorial about the standard, and a MP4-SA to C translator that produces runtimes that can work in real-time. It's a "meta-answer" to the Ask Slashdot question; it's probably not going to help today, but I think its the right techology for Linux to build on.
This topic reminds me of the middle ages, when it was thought commoners just couldn't be expected to learn how to read. So elaborate art was commissioned to teach the Bible via pictures, and the rest was handled through the oral tradition.
Oh, and not teaching the commoners to read had nice side effects for the upper classes too. Then came Gutenberg, and the next thing you know you Tony Blair is tossing out the inherited members of the House of Lords :-).
Fast forward to the 21 century, we see history replaying itself -- now commoners are considered too stupid to program instead of too stupid to read. Having watched secretaries with high-school educations become TeX Macro wizards, I can say the commoners are pretty damned smart, and are starting to notice they have nothing to lose but their chains.
If your application can be fielded entirely in ANSI C and its libraries, you can just use dos-gcc to compile up a DOS executable under Linux (dos-gcc is djgpp built as a cross-compiler). I build each sfront release under Linux for both Linux and DOS/Win32 using dos-gcc, and it works great. It was worth purging gratiutious non-ANSI-libs from the source tree ...
Some of us developed Open-Source UNIX console apps that are strict ANSI-C. Without leaving the comfort of our UNIX machines, we can use dos-gcc to make .exe files to distribute along with our tar.gz's and rpm's.
My question to you is, how do you get the rest of the "culture" right for the DOS distro, if you're unwilling to immerse yourself in DOS by using it? I'm basically talking about doing all the stuff that "rpm -i" does automatically on a Linux box, along with issues like portable uses of the ANSI system() and getenv() to do things like find a local C compiler and execute it from your C program. Is there a HOW-TO or FAQ about making your UNIX console program fit (as opposed to work) in a DOS world? Thanks in advance!
P.S. To put the question in perspective, the app I actually maintain is a program that generates C files, thus the compilation issue.
When I was a grad student at Caltech in the 80's, we used MOSIS (www.mosis.edu) as a silicon broker to fab IC chips. It was all done via the Internet, we sent CIF design files via email, and automatic responders sent update messages tracking the chips progress. Three months later, a Fedex box would show up with the test chips, and a goverment grant (and before the 80's ended, a VISA card of a company or individual) got billed. So, I'd say Internet shopping worked quite well pre-PC, as long as you were buying IC fab services :-).
At AES last year I went to a Dobly demo room, where they remixed different types of music for 5:1. Musically, I can't say it added anything to the experience for me.
Now, for movies and video games its different, placement around the space has meaning in the context of the story, but there's a long tradition of musicians all being in one place -- on a stage -- and people sitting in front of them, which is a good match for stereo placement.
Nesheim, John L., 1942- High tech startup : the complete how-to handbook for creating successful new high tech companies / John L. Nesheim. Saratoga, CA : Electronic Trends Publications, c1992.
Found in good university bookstores (Stanford bookstore has it), or at amazon.com
Analog neural nets work best in special purpose application -- where what you want to do is to take a stream of input vectors, feed it into a MLP, and generate output vectors, ideally at the maximum rate that the analog circuits work. Or, if you're into power instead of speed, push down the power used by the net until you're just barely making the timing. This works the best when there's integrated sensors (like photoreceptors or a silicon cochlea) or actuators (micromachined stuff) on the chip, so you dispense with an A/D or a D/A by going analog as well.
There are a few products out there that use analog nets just in this way -- there's one in your Logitech Marble Trackball, computing the motion vectors of a pseudo-random dot pattern on the ball in IR.
But most problems out there actually don't fit in that nice little niche -- instead, the neural net is part of a signal processing chain, and the adjacent steps don't fit well in analog, so you need a DSP anyways. And once you have the DSP, its usually is a lose to put an analog neural net on the die, instead of just adding another ALU to the DSP, since the rest of the system is digital anyways.
One hope is this networking protocol, the address-event representation, that lets analog neural net chips communicate with each other in a very digital way, while being very efficient to implement in analog. See this paper for details ...
I think its important to remember our ... I know when we put out
... is just that -- putting out nice ...
roots here
our E-CAD package under the GPL in 1990
www.cs.berkeley.edu/~lazzaro/chipmunk
we weren't doing it for "world domination",
we were doing it so there would be nice
(for the time) tools for people to use
to design chips. And from what I can tell,
a lot of the spirit of gcc, emacs, Linux,
ect
tools by and for a community. If we keep
the focus on quality and community, and
not the folks at Redmond, only good things
can happen
How do killer apps happen? To use Geoffry Moore's
analogy, they are the result of the Tornado --
technology and users needs coming together at
just the right point in time, so that a lazy
product segment becomes an unstoppable force.
And in the end, one company ("the gorilla")
dominates the segment. Looking back in history,
relational databases were a tornado, Oracle
became the gorilla. Internet infrastructure was
a tornado, Cisco became the gorilla. Desktop
publishing was a tornado, and the Macintosh and
Adobe shared joint gorilla status until recently.
On the free software server front, Apache and
Samba were the real gorillas, and Linux came
along as infrastructure for them. If we're
looking for desktop "killer apps", we need to
look to the future, predict the most probable
Tornados, and be ready -- not look in the rear
view mirror at Office, Microsoft got there first.
Especially since most of us don't even use an
app like Office to get our own work done -- we
use TeX+emacs.
> Keeping a company private doesn't
> do much to the share price.
In practice, if an IBM carries the stock of
a small public company on their books, it's
not reflected in IBM's stock price in a
material way. And if an IBM sells the stock, its
seen as a "one-time gain", unimportant in
analyst's eyes. Strategic reasons dominate in
investments like this, and an IBM's preference for
RedHat going public or staying private lies in
strategic issues, not bottom-line issues.
> Support is a (sic) primarily a loss leader.
Take a look at IBM's last quarterly report --
services provide the lion's share of earnings.
> What can RedHat add?
Coca-Cola == RedHat
RC Cola == Cheapbytes RedHat distribution
In the end, cola is cola -- why do the vast
majority of people walk past the RC Cola and
buy the more expensive Coke? Brands are powerful
things. RedHat is setting themselves up to be
the 21st century Apple -- building a brand that
means high-tech rebellion, just like Apple did.
But what does Ben Rosen know about managing a ... just like
computer company where you sometimes have to
design the key components (hardware and software)
yourself? It's like taking the CEO of Rite-Aid
drug stores and putting him in charge of Merck.
And if Rosen doesn't get it, the person he hires
to take over probably won't either
Pfieffer didn't.
There never really was much synergy ...
with buying Tandem and DEC, it harked
back to an era when CEO's didn't get
humunous stock options, and so they
built empires to gratify their egos.
Watch for a trivestiture with the new
CEO
> The problem is that we have presented no
...
> alternative economic model for companies
> who truly want the advantages of the bazaar
> --but still have bills to pay.
It's imporant to note that several types of
occupations, from airline pilot to modern
dancer, have become harder to justify doing
in purely economic terms, because there are
enough people who want to do them who aren't
primarily concerned about total monetary
reward amortized over the career. As a result,
people for who the "bills to pay" issue is
significant simply do something else for a
living. Maybe a part of what we're seeing here
is certain types of computer programming entering
this territory