TRIBINA "BUDUĆNOST VEŠTAČKE INTELIGENCIJE"

[zakk]

grozno -_-

[mac]

Ljudi su počeli da prepoznaju tekstove generisane pomoću AI. Louis Rossman (jutjuber koji se bori za consumere, što zapravo nije bitno za ovu priču) ima SKILLS md fajl koji može da vam pomogne u vezi s tim.

https://github.com/realrossmanngroup/no_ai_slop_writing_rules/tree/main

Ima dva SKILLS fajla, jedan je taj no-ai-slop, a drugi je rossmann-voice, koji omogućava LLM-u da imitira samog Rossmana.

PS. Prvo sam stavio tačku između SKILLS i md, ali .md je domen za Moldaviju, pa se to pretvori u link. U AI eri Moldavci bi mogli lepo da zarade od samo tog jednog sajta.

[Onaj stari Sendmen]

Pa dobro, ali opet koristiš AI, tj. Claude Code da prepoznaš "ai slop" koristeći taj md fajl. Missim, how this helps? This is just more of going down the same rabbit hole. 

[Meho Krljic]

Sam Tucker stavlja u komični geg poslednjih šest meseci AI-pokalipse:


If AI Had to Pass Job Interviews:

https://www.youtube.com/watch?v=5TYB6uy4w7M

[Васа С. Тајчић]

Ако се програми за претварање звука у текст као што је Whisper рачунају у вештачку интелигенцију, онда сам морао да се дебело црвеним пред неким људима.

[Meho Krljic]

Pa jeste to neki AI, ali ljudi danas uglavnom misle na LLM-ive i generativni softver kada pričaju o veštačkoj inteligenciji, tako da, ajde da kažemo da si bezbedan za sada.

[Meho Krljic]

Mo Bitar je sve zabavniji poslednjih dana dok se divlje rve sa sve filozofskijim pitanjima. U najnovijem videu od pre par sati, a kome je promenio naslov da bi bio manje kontroverzan (prvo se zvao "I now know why AGI doesn't exist") izgovara par rečenica koje su verovatno suština:

"The reason we can'tr build intelligence with machines is because intelligence doesn't exist abstractly in the real world. It's not something that the human species models. It's something we are."


Danas sam baš pisao o Tedu Chiangu i njegovoj veličanstveno noveli The Lifecycle of Software Objects a u kojoj, da parafraziram deo svoj prikaza (izlazi prekosutra), Chiang kredibilno pokazuje da inteligencija nije stvar koja se dizajnira i da je ona ishod iskustva, to jest da se ne može napraviti prečica do nje. Pa me sve ovo posebno slatko pogodilo.

Enivej, Mo:


https://www.youtube.com/watch?v=2bVS18lhTEA

[mac]

Jeste, treba iskustvo, ali ne mora da bude sopstveno. Uostalom ako mašina usvoji kao sopstveno iskustvo koje dobije onda će to zaista biti iskustvo mašine.

Trenutni LLM modeli ne pružaju te mogućnosti, ali radi se na novim tipovima modela. Jan LeCun radi na takozvanom "world model". Ni to neće biti poslednja reč na tu temu, ali biće bolje od ovog što trenutno imamo.

[mac]

Berni Sanders hoće novi zakon za AI

Introducing the American AI Sovereign Wealth Fund Act

[Meho Krljic]

Dokinsova zabluda o Klodu

[Meho Krljic]

Inače, koga mrzi da se detaljno sam informiše o guglovom najavljenom pivotu na AI pretragu i zašto je to strašno, El ima tipično metodično istražen i artikulisano isporučen video:

https://youtu.be/jQyKd1_e3Xg

[Meho Krljic]

Mo Bitar nastavlja da briljira. Današnji video nije specijalno filozofski nastojen, ali sažima stvari vrlo dobro. Ko nema vremena za JAKO DUGAČKE njuzletere i intervjue koje radi Ed Zitron o neprofitabilnosti AI industrije, ovde ima isto to u probavljivih, zabavnih osam minuta.



"80% of the time you're listening to confident nonsense that sounds like a stoner explaining quantum physics" je najsažetiji siže LLM-ova do sada poznat ljudskoj rasi.


https://www.youtube.com/watch?v=lwjVjD3oQJg

[Meho Krljic]

Džejk (Mancarryingthing) u kratkom skeču o "neizbežnosti" AI:

https://www.youtube.com/watch?v=M0-kPte_Erc

[Meho Krljic]

Is AI Slop Worse Than Cannibalism?


https://www.youtube.com/watch?v=YCPAIg7RUq8


Svi znamo odgovor bez gledanja, ali je ipak zabavno.

Truman posebno da pogleda.

[mac]

Uf, ima par ljudi na fejsbuku kojima bih rado pokazao ovo, ali naljutili bi se što ih poredim s kanibalima...

[mac]

Mada, potpuno isti argument može da se iznese i za pušenje, drogiranje, alkoholisanje, kockanje, pa čak i konzumiranje porno sadržaja. Pa ajd sad upotrebi nešto od navedenog da ubediš ljude da je AI loš. Ako nisi u muslimanskoj državi džaba trošiš dah.

[Onaj stari Sendmen]

Why big AI labs are hiring so many philosophers

The technology presents all sorts of thorny problems—a philosopher's favourite kind

This rise in recruitment of philosophers at AI companies is a result of the broader understanding that philosophy offers useful tools for AI development.

Why Big AI Labs Are Hiring More Philosophers? There is a belief that AI training may be able to benefit from some of the ancient lessons of philosophy.
The rise of the artificial intelligence era is changing expectations about careers. While earlier, there were concerns that arts and humanities subjects would be more impacted, the trend appears to have shifted.

According to The Economist, AI companies are actively recruiting philosophers, sometimes even before graduation. This rise in recruitment of philosophers at AI companies is a result of the broader understanding that philosophy offers useful tools for AI development. There is a belief that AI training may be able to benefit from some of the ancient lessons of philosophy.

For instance, the Socratic method, developed in ancient Greece by Plato, encourages structured questioning to test ideas and uncover contradictions. According to The Economist, researcher Jörg Noller argued that this approach can help make AI systems less agreeable and more focused on truth, rather than simply pleasing users.

Another similar idea is that of the "Socratic ignorance." In Plato's 'Apology', Socrates says that true wisdom lies in knowing how little one knows. Experts are of the view that building this humility into AI systems could reduce their overconfidence. Iason Gabriel, a senior philosopher at Google DeepMind, an AI lab based in London, attributed an industry-wide decline in hallucinations to such efforts.

Philosophy can also shape how AI systems behave. Thomas Powers of the University of Delaware has argued that if an AI legal assistant is trained on John Locke's writings, it may place strong emphasis on property rights as part of political freedom.

The report added that some developers even allow users to tune such values. For instance, the "Granite" series of models from IBM, an American computing giant, includes settings that let companies adjust outputs to match different corporate philosophies.

At the same time, philosophy can also be helpful in promoting safety in AI tools. To prevent harmful actions like deception or blackmail, some systems are using "AI constitutionalism", that are based on moral and legal principles. Anthropic has built constitutions for its Claude models using ideas from Immanuel Kant, Apple's terms of service and the Universal Declaration of Human Rights.

There is also a debate about what moral rules should guide AI systems. According to The Economist report, one major approach is deontology, which is linked to Immanuel Kant. This idea says AI should follow strict rules like not lying, not coercing people, and not using people just as tools, even if breaking the rules might produce a "better" outcome. Some AI labs, like Anthropic, include these kinds of rules in their systems.

"These can make AI behaviour more consistent....a plus for deploying robots in homes and public spaces," Dr Powers told The Economist.

The second ethical approach for AI is "consequentialism," which focuses on choosing whatever produces the best overall results. This approach is used in systems like ChatGPT and Gemini, the report added.

While AI models may increasingly be tuned to these philosophies in the coming years, critics also worry about "moral deskilling," where people lose the ability to make ethical decisions if AI makes them for them.

"Morality is historically unstable, culturally variable, strategically manipulable, and often only retrospectively legible," Roman Yampolskiy, an AI theoretician at the University of Louisville, argued, noting that this makes ethical decision-making difficult to formalise in AI.