Anthropic je zgradil demokratičnega klepetalnega robota z umetno inteligenco tako, da je uporabnikom omogočil glasovanje za njegove vrednote

Anthropic je zgradil demokratičnega klepetalnega robota z umetno inteligenco tako, da je uporabnikom omogočil glasovanje za njegove vrednote

Časovni žig: 18. oktober 2023 1

In what may be a first of its kind study, artificial intelligence (AI) firm Anthropic has developed a large language model (LLM) that’s been fine-tuned for value judgments by its user community.

Veliko javnih LLM-jev je bilo razvitih z zaščitnimi ograjami – kodiranimi navodili, ki narekujejo specifično vedenje – v poskusu omejitve neželenih rezultatov. Anthropic's Claude in OpenAI's ChatGPT, na primer, uporabnikom običajno nudita pripravljen varnostni odziv na izhodne zahteve, povezane z nasilnimi ali kontroverznimi temami.

However, as innumerable pundits have pointed out, guardrails and other interventional techniques can serve to rob users of their agency. What’s considered acceptable isn’t always useful, and what’s considered useful isn’t always acceptable. And definitions for morality or value-based judgments can vary between cultures, populaces, and periods of time.

Povezano: Združeno kraljestvo se bo na načrtovanem novembrskem vrhu osredotočilo na potencialne grožnje z umetno inteligenco

One possible remedy to this is to allow users to dictate value alignment for AI models. Anthropic’s “Collective Constitutional AI” experiment is a stab at this “messy challenge.”

Anthropic je v sodelovanju s Polisom in Collective Intelligence Project izbral 1,000 uporabnikov iz različnih demografskih skupin in jih prosil, naj odgovorijo na vrsto vprašanj z glasovanjem.

Anthropic built a democratic AI chatbot by letting users vote for its values PlatoBlockchain Data Intelligence. Vertical Search. Ai.
Vir, Antropično

Izziv je osredotočen na to, da uporabnikom omogočite agenciji, da ugotovijo, kaj je primerno, ne da bi jih izpostavili neprimernim rezultatom. To je vključevalo pridobivanje uporabniških vrednosti in nato implementacijo teh idej v model, ki je že bil usposobljen.

Anthropic uses a method called “Constitutional AI” to neposredna its efforts at tuning LLMs for safety and usefulness. Essentially, this involves giving the model a list of rules it must abide by and then training it to implement those rules throughout its process, much like a constitution serves as the core document for governance in many nations.

In the Collective Constitutional AI experiment, Anthropic attempted to integrate group-based feedback into the model’s constitution. The results, po to a blog post from Anthropic, appear to have been a scientific success in that it illuminated further challenges towards achieving the goal of allowing the users of an LLM product to determine their collective values.

Ena od težav, ki jo je morala premagati ekipa, je bila izdelava nove metode za postopek primerjalne analize. Ker se zdi, da je ta poskus prvi te vrste in se opira na metodologijo ustavne umetne inteligence podjetja Anthropic, ni uveljavljenega testa za primerjavo osnovnih modelov s tistimi, ki so prilagojeni vrednotam množice.

Ultimately, it appears as though the model that implemented data resulting from user polling feedback outperformed the base model “slightly” in the area of biased outputs.

Glede na objavo v blogu:

»Bolj kot nad nastalim modelom smo navdušeni nad procesom. Verjamemo, da je to morda eden prvih primerov, ko so člani javnosti kot skupina namenoma usmerjali vedenje velikega jezikovnega modela. Upamo, da bodo skupnosti po vsem svetu gradile na takšnih tehnikah za usposabljanje kulturno in kontekstno specifičnih modelov, ki bodo služili njihovim potrebam.«

Časovni žig:

Več od Cointelegraph