Agents are still too expensive. What we can do, ultimately, is build tools that agents can easily use.
When the era comes where the general public uses agents, they'll run on cheap, lightweight Chinese open-source models.
We need to build the tools now so those models can use them β tools so simple that even the basis for an agent's own judgment can be retrieved with a single query.
μμ΄μ νΈλ μ¬μ ν λ무 λΉμΈλ€. κ²°κ΅ μ°λ¦¬κ° ν μ μλ 건, μμ΄μ νΈκ° μ½κ² μ¬μ©ν μ μλ λꡬλ₯Ό λ§λλ κ²μ΄λ€.
λμ€μ΄ μμ΄μ νΈλ₯Ό μ¬μ©νλ μλκ° μ€λ©΄, μ¬λλ€μ μ λ ΄νκ³ κ°λ²Όμ΄ μ€κ΅μ° μ€νμμ€ λͺ¨λΈμ μ°κ² λ κ²μ΄λ€. μ§κΈ κ·Έ λͺ¨λΈλ€μ΄ μ¬μ©ν μ μλλ‘ λꡬλ₯Ό λ§λ€μ΄λμ΄μΌ νλ€.
μμ΄μ νΈκ° μ€μ€λ‘ νλ¨ν μ μλ κ·Όκ±°μ‘°μ°¨ κ°λ¨ν μ‘°νν μ μλ, κ·Έλ° λꡬλ€.