IBM superračunalnik simulira mačje možgane

Zakaj bi pa moral sploh pain zdrzat?

No pain, no gain. Ne znaš cenit drugače

O.

In kaj če ti vsadijo pain v spomin, brez da bi ga doživljal? :)

To je itak jasno... maš večnost pleasurja, če zdržiš 1urco max pain. Zraven imaš pa gumb, ki prekine pogodbo.

Odin ti nebi zdržal niti 10sekund.

Eh, brezveze je diskutirat koliko bi kdorkoli vzdzal, ker ta scenarij nikoli ne bi prisel v postev.
Enacba qualije je tisto kar potrebujemo, da si lahko "predstavljamo" kako bi Protokol zgledal.
In ta enacba bi potem tudi pojasnila ali se splaca reskirat pain (v primeru, da gre nekaj narobe in se ustvari Hell SAI), ali ne.
Na to bi morali gledat bolj objektivno in matematicno, ne pa kaj bi kdo zdrzal.

Recimo:
enota qualije = q
q = 0 - no pleasure, no pain
q > 0 - pleasure
q < 0 - pain
E - energija

Ce je q=E, potem bi se lahko mirno sami odlocili kaksno razmerje hocemo med ekstazo in casom.
Ce je recimo q=E², potem pa bi kurili Zemljo pocasi googol let z minimalnim pleasurjem.
Ce pa je q²=E, potem pa bi morali skurit Zeljmo cimbolj intenzivno in hitreje.

Ce je "maxpleasure + maxpain = 0", potem se spaca reskirat moznostm 10% Hell SAI.
Ce pa je, kot pravi Double_J, "maxpleasure + maxpain" < 0, potem pa ne.

Ja, enacbo rabimo.. Ampak je bo tezko dobit brez SAI.

Ja no sej, lahko damo bolj življenske primere.

Če bi ti kdo ponudil hud in bogat lajf, v zameno da te vsako nedeljo sežge.
Kaj pa je to tistih 10minut proti enem tednu... ampak ne štima tko.

Sej rizik je itak majhen za hell in nimamo posebnega vpliva nanj. Nekaj varovalk se ugradi, kakšen scenarij konkretno, se pa odvija pa pomoje ni posebaj izrazit faktor nevarnosti.

Cela Drama, lol.

Hehe, da se ne bo Protokol spremnil v Hell machine... že vidim scenarij, kako zlobni hekerji preprogramirajo zadevo...

Bolj mogoč kot LHC doom, enako, ali manj? Konkreten izračun ali občutek?

I am not going to take sides in this debate, and I definitely agree that the IBM claims are probably outrageous, but there are several aspects that might lead to a more "fair and balanced" thinking about the whole issue: - first, we do not know how much the results from the IBM researchers have been exaggerated by the news/PR people. I know situations in the past when PR people have turned reasonable research results into outrageous earth-shattering bogus claims. I don't have indications that it was the case here, but I would not be surprised if this was a factor that exaggerated the magnitude of the claims. - second, I'm not sure that Henry Markram's argument is 100% bullet-proof. While I am not a neuroscientist, I am a circuit designer, and I know first hand the "power" of hierarchy and abstraction. In a modern integrated circuit there are billions of transistors, which looked at individually are quite complex, they have 3D phenomena galore, in order to understand them in detail you need to use quantum mechanics, solve Schroedinger's equation, etc., yet we can use much simpler "compact models" for the transistors, and then even higher level behavioral models for the functional blocks, to design and simulate very complex structures (e.g. a microprocessor), that would be unfeasible if we insisted on the detailed simulation of each aspect of the transistor behavior. This way we lose a little bit of accuracy, but can get most of the important behavior. I think this is the premise behind what IBM is trying to do, and while I don't know that it should work in the case of neural networks, I am not aware of really strong arguments that it should not. - finally, there are indications that having all the complexity of neural models advocated by Henry Markram may actually not lead to better predictive results (to say the least). In a recent competition between what I would call "compact models" (aka threshold-based) and detailed models (e.g. Hodkin-Huxley), the compact models won the race by "fiat", see: http://www.incf.org/community/competiti... Science 16 October 2009: Vol. 326. no. 5951, pp. 379 - 380, "How Good Are Neuron Models?" by Wulfram Gerstner and Richard Naud Hope this helps! Cheers, Mircea .

En dober komentar, na novico o "cat fightu".

Lepo pokaže/pove, zakaj se mu zdi da MArkram pretirava. Se popolnoma strinjam.

Hehe, se pravi so eni malce pretiravali v svojih "teorijah". As usual, bi rekel.

Saj verjetno nihče ne dvomi, da bomo nekoč znali narediti elektronske možgane. Vprašanje je, kako daleč smo v resnici že napredovali. Markham je po mojem kar prav povedal, da ta simulacija niti dobra simulacija mravljinčjih možganov ni, kaj šele mačjih. Če drugega ne, izhaja trditev o simulaciji mačjih možgan iz nedvomno napačne predpostavke, da je v možganih kompleksnost sama pomembnejša od oblike. Način povezav je v živčnem omrežju precej pomembnejši od števila povezav, o načinu povezav pa ne vemo še praktično ničesar.

O.