×

Announcing: Slashdot Deals - Explore geek apps, games, gadgets and more. (what is this?)

Thank you!

We are sorry to see you leave - Beta is different and we value the time you took to try it out. Before you decide to go, please take a look at some value-adds for Beta and learn more about it. Thank you for reading Slashdot, and for making the site better!

Comments

top

Turing Test Passed

Olivier Galibert Re:Searl missed the point. (432 comments)

We are at the point where a computer can read a novel and spit out a high school book report that would both fool and impress most english teachers, and it can do it in seconds not days.

Not quite. It's very possible to do things that work part of the time, and allows for very nice demos. But the systems very easily blow a gasket on wrong parses, out-of-domain knowledge, etc. Roughly, there are three problems: we don't know how to operationally represent meaning, we don't know how to handle concepts that are fuzzy around the edges, which is the case of pretty much every concept out there, and we don't know how to introduce in a system all of the world knowledge a normal adult has.

Note that the advent of magnificient things like wikipedia certainly help, but as far as I know nobody is able to bootstrap a system from it yet.

There are also a lot of posts claiming the Turing test doesn't mean anything. However none of them I have read so far actually explain their statement, so I assume they are parroting their philosophy proffessor who was probably referring to Searle's Chinese translation room argument.

If you ever work in dialogue systems, you'll find out how adaptive humans are in a communicative context. It's, in fact, relatively easy to push a human to say things a particular way your system handles better, and he won't even notice. And that's because humans do it all the time. It's not a bad thing at all, and makes building efficient dialog systems for real tasks a tad easier. But it can shift the focus of the turing test from answering like a human to fooling a human, which is not the same problem at all and, annoyingly, a way easier and less interesting one.

    OG.

about 6 months ago
top

Fusion Reactor Breaks Even

Olivier Galibert Re:Here's the real story (429 comments)

Weapons research is the only place where you get enough funding for a long enough time to througly do the research work. So you get results.

    OG.

about a year ago
top

Apple Hacker Charlie Miller To Demo Dangers of Near-Field Communications

Olivier Galibert Re:What makes you think his "sentence" is ever up? (149 comments)

Well, the windows 8 upgrade cost is announced to be $40, so it's not considerably more.

And yes, $29.99 is upgrade since you're only allowed to use it on mac hardware, and such hardware is always sold with a copy of osx.

    OG.

more than 2 years ago
top

Apple Is Forced By EU To Give 2 Years Warranty On All Its Products

Olivier Galibert Re:As An American... (270 comments)

This is standard consumer protection stuff. Does the US have a directly equivalent law? No idea, but it doesn't lack laws that are in the same ballpark. Indeed, some, such as the requirement that all electronics be vetted by the FCC and contain shielding to prevent their circuits from accidentally broadcasting something that might cause a little interference on a TV or radio in the same room, seem a tad less understandable than creating a basic standard of merchantability - you have to stand behind your product for two years. Hardly unreasonable.

What gives?

Errr, Europe has the same non-interference/resistance to interference laws that the US have.

    OG.

more than 2 years ago
top

T-Mobile Exec Calls For End To Cell Phone Subsidies

Olivier Galibert Re:There needs to be a way to avoid the subsidy. (355 comments)

Wow, and here in France for $27/month we get unlimited voice, unlimited text and 3Gb of data. And you can stop whenever you want without cost. You guys are really getting fucked sideways.

    OG.

more than 2 years ago
top

Google Asks Court Not To Enjoin ReDigi

Olivier Galibert Re:Can the courts decide A = !A (185 comments)

That's what the appeal process and supreme courts are for. There's only one supreme court per state (for state issues) and one federal supreme court so that the final saying is, well, final and non-contradictory.

    OG.

more than 2 years ago
top

Chinese Government Ramps Up Weather Control Efforts

Olivier Galibert Re:Climate change... (139 comments)

That's where it can get interesting... Water vapor is a greenhouse gas, a *way* more efficient one than CO2. If they manage to reduce the amount of water vapor in the atmosphere, that is going to reduce the greenhouse effect.

Interesting, 'innit ?

    OG.

more than 2 years ago
top

Google To Seek Dismissal of Suit Against Google Books

Olivier Galibert Re:thats simply wrong (240 comments)

No it isn't. A full reproduction can be fair use. What you do with it is a large part of the definition.

    OG.

more than 2 years ago
top

When Algorithms Control the World

Olivier Galibert Re:I guess doom and gloom sells more ads (150 comments)

The first Ariane 5 exploded on launch because a feedback mechanism for guidance had a sign swapped, again creating positive feedback.

That's incorrect. The first ariane 5 exploded because of correct, reused ariane 4 code becoming incorrect in the new environment. More specifically steering code which results are used at the start of the flight and unused but left running afterwards. The code was still correct in start-of-flight conditions, but in the afterwards condition noticed speeds way over what it was supposed to see and triggered a security abort (ariane 5 is a tad faster than ariane 4).

So no sign errors, no feedback, just correct code running at a time it shouldn't have and untested there.

    OG.

more than 3 years ago
top

Test Driving GNU Hurd, With Benchmarks Against Linux

Olivier Galibert Re:GNU vs GNU (335 comments)

The differences are going to come from the memory management, i/o paths, disk i/o speed/scheduling, etc. Which is what the kernel is about, after all.

Too bad Hurd is still so limited that you can't really run it on real hardware, otherwise the benchmark differences would probably be much worse...

    OG.

more than 3 years ago
top

"Space Archeology" Uncovers Lost Pyramids

Olivier Galibert Re:No imagination... (156 comments)

That's called "Alien vs. Predator".

    OG.

more than 3 years ago
top

Artificial Leaf Could Provide Cheap Energy

Olivier Galibert Re:5.5% of the energy in sunlight into hydrogen fu (326 comments)

One of the numerous problems with hydrogen is that you need very high pressures to store any decent amount of it in a container. And anything at very high pressure has the protential to be extremely dangerous. So more dangerous in handling stored than natural gas. As a transfer medium or as buoyancy though, I agree.

    OG.

more than 3 years ago
top

Best-Selling Author Refuses $500k; Self-Publishes Instead

Olivier Galibert Re:FYI (290 comments)

I even happen to know that. Sleep, I need more of.

    OG.

more than 3 years ago
top

Best-Selling Author Refuses $500k; Self-Publishes Instead

Olivier Galibert Re:Was wondering when this would happen (290 comments)

Out of the $20, there's approximatively $6-8 for the library (amazon being the worst there, they want $10 iirc), $3-4 for printing, $2-4 for shipping and the distributor if there is one (and in NG's case, I'm sure there is). So that's $4-9 left, i.e. $2-7 for the publisher (and closer to $2 than to $7). The publisher is definitively not ripping him off.

    OG.

more than 3 years ago
top

Intel Unveils Next Gen Itanium Processor

Olivier Galibert Re:Isn't it strange... (169 comments)

That's because the high-end server world accepts level of single-core performance the consumer world doesn't. These processors are not something you want on your PC. You want something with better memory management, way faster I/O with ram and GPU, etc. OTOH, you usually don't care about multi-processor.

But faster I/O usually means putting more things on the die (hence amd's integrated memory controllers, now followed by Intel) and having larger busses/more efficient protocols, and acting on that means changing the socket. And the north bridge, if one is left. And the memory, for a faster one. You wouldn't get enough speedup from changing the cpu alone with everything else pin-compatible to make it worth it.

Meanwhile, the itanic spends its time waiting for the ram to answer... but since you put a lot of them in the box, in aggregate they can be useful.

    OG.

more than 3 years ago
top

Stanford, UCD Researchers Say 100% Renewable Energy Possible By 2050

Olivier Galibert Re:PR Puff Piece (360 comments)

Niiiiiiice. $19 trillions just for the wind turbines (around 5M each), $100 trillions for the rooftop PV systems (around 60K each), but there is no economic issue. Right.

Only $135 billions for the dams (around 500M each)... if you can find 270 new places in where to put them...

    OG.

more than 3 years ago
top

Why Linux Loses Out On Hardware Acceleration In Firefox

Olivier Galibert Re:modularity (456 comments)

And fixing that is probably not going to ever happen until X/Mesa is dead under its own weight.

X is a protocol, and a pretty good one at that. There is no reason for it to "die", since nobody has come up with anything better yet. Both the Windows and the OS X graphics architectures are inferior.

The X server software and Mesa should get updated. But it actually works pretty well. Most of the things you list are fairly specific add-ons, and having those access the hardware separately seems like a good thing; why would I want to have all that extra crap in a single project?

People need to do some refactoring, cleanup, and documentation. But, hey, what else is new. But there is nothing really wrong with having those different pieces of functionality factored into seperate projects.

X11 the protocol was very good 20 years ago, but by now shows its age. A new X12 could use some cleanups such as removing colors, palettes and visuals (truecolor is the only relevant one nowadays), adding explicit gamma support, removing every drawing primitive except unaliased points, horizontal/vertical lines and image handling, adding alpha support, adding efficient image transfer mechanisms by mapping video-card accessible buffers in the application (XShmPutImage and glTexSubImage2D are sad jokes, performance-wise), adding blending/compositing support, etc. And that's just graphics, don't get me started about multi-screne handler, internationalization, window management or inter-process communications.

As for the "specific add-ons", I was talking about 2D X rendering, you know, the thing that draws your windows, 3D rendering, which you may have heard about, and video decoding, the thing linux users do in software because less that one "standard" per video card vendor would be unacceptable. Vaapi happens to be intel's (vdpau is nvidia's, xvba is amd's). Nothing obscure there. And if you think they're independant you either haven't looked at the problem and the hardware or your blood level is too high in your coffee stream.

At first sight, the opengl 3 level intel cards have three beautifully separated subsystems, 2D blitter, 3D renderer, media decoder. Then you find out that only one can be used at a given time, and you have to explicitely switch between them. And, in addition, things like glClear are better handled with the 2D blitter, compositing, frame filtering and deinterlacing with the 3D hardware, etc. And in practice you want to unify pixmaps, textures and movie frame buffers, otherwise pain ensues when you want to use the damn things. So the interdependance level is actually high. As a result it is *very* wrong to have these pieces in separate projects, because communication layers and version issues multiply exponentially.

Finally, I don't know about OS X, but the Windows architecture for video drivers is actually superior than the current linux ones. You have one kernel-level driver and one userspace driver, and that's it. The API is not the best possible by far, with way too many functions, communication paths and a somewhat obsolete shader microcode. But the unification of all the hardware functions in two drivers with the kernel boundary in between is the best you can have.

And, the point you missed, is that the refactoring *will* *not* *happen* for political and social reasons. Even before patches, the first step would be to unify all that stuff under one tree, and the pushback against that is demented.

    OG.

more than 3 years ago
top

Why Linux Loses Out On Hardware Acceleration In Firefox

Olivier Galibert Re:The state of the graphics stack doesn't help (456 comments)

X11 is the problem. People will really start looking at it when OS X releases full OpenGL 3.x across the entire user space for OS X [Linux isn't even at 2.x and Windows doesn't use it for its DE drawing environment--just app specific]. When will people start calling X11 long in the tooth and needs to be completely replaced?

Oh please, when you bitch about something use real facts. Mesa is at OpenGL 2.1, and the, I think, hardest part for reaching 3.x, a full featured optimizing GLSL compiler, is well on its way. As for Windows, the DE hits the 2D and 3D parts of the DirectX driver, obviously.

    OG.

more than 3 years ago
top

Why Linux Loses Out On Hardware Acceleration In Firefox

Olivier Galibert The state of the graphics stack doesn't help (456 comments)

Graphics drivers are all over the place. For instance, the intel stack, to be complete, requires:
- the xserver tree
- the protocols tree
- the libdrm tree
- the intel 2d video driver (includes separated DDX driver and XvMC driver)
- the kernel (drm tree)
- mesa with its integrated drivers
- libva (for vaapi)

That's 5 hardware-accessing drivers (internal kernel, DDX, XvMC, internal Mesa, libva) in 4 trees linked together with libraries and applications coming from 3 more trees. And they call each other through layers and layers of function arrays with no real documentation at any level. It's always fun when trying to understand a function to see it calling another one through a function pointer which after two more indirections finally ends up in another function a paragraph after the original one. And you have to trace everything, because the just as innocuous call after that one is in fact going to send a message through a drm connection and the X server to the DDX driver. And will be as documented as the previous one. Add to that a (failed, but present) tentative in the code to support almost any combination of versions in this dreadful house of cards, and you end up with an astounding amount of added complexity that does not make debugging easy.

And fixing that is probably not going to ever happen until X/Mesa is dead under its own weight. The bitching when the n protocol trees became the one protocols tree was incredible, I don't see the poor soul who managed that one doing it ever again.

    OG.

more than 3 years ago

Submissions

Olivier Galibert hasn't submitted any stories.

Journals

Olivier Galibert has no journal entries.

Slashdot Login

Need an Account?

Forgot your password?