Enrique

[Enrique @Enrique_v_g]

@RulinShao @shannonzshen @oscar_yinn @yuetai12575 @hamishivi @natolambert @TengX6 @ml_perception @scottyih @LukeZettlemoyer @PangWeiKoh What about costs? Won’t this be a cache nightmare?

[Enrique @Enrique_v_g]

@giadotai Cache

[Enrique @Enrique_v_g]

@thsottiaux I trusted you and burned my tokens. Now I have actual important stuff I can’t finish 💩

[Enrique @Enrique_v_g]

@SwapnanilDhol @OpenAIDevs @thsottiaux Same, from 44% to 0

[Enrique @Enrique_v_g]

@avi_eisen 50%

[Enrique @Enrique_v_g]

@chacon Compared to inkling, which probably has orders more of backing resources…

[Enrique @Enrique_v_g]

@analogalok Looks like benchmaxxing (or a flawed benchmark or both)

[Enrique @Enrique_v_g]

@jenzhuscott Privacy matters

[Enrique @Enrique_v_g]

@samhogan @inference_net Why didn’t you use deepseek pro, and just use flash?

[Enrique @Enrique_v_g]

@BrendanFalk @pmarca Don’t give access to api keys and passwords to an agent…

In this case “Three clocks bloom at midnight” is similar to prompt injection (latent injection), assume your agents will be prompt
Injected, understand the attack surface and add human in the middle accordingly