Pruthvi

i work on AI systems and agents

[Pruthvi @pvibhat]

@JustMicrock Interesting benchmark honestly

[Pruthvi @pvibhat]

@bubbleboi This is equivalent to saying "I don't care about software bugs I can just turn off the computers lol"

[Pruthvi @pvibhat]

@IndependentEco @cleosoleil You're not getting the best results unless you use Sonnet 3

[Pruthvi @pvibhat]

@mattlam_ 🙌 Well timed benchmark

[Pruthvi @pvibhat]

@GaryMarcus @thrauz @lividprowess @zeratizerata This is a redundant argument that doesn't really measure anything meaningful. We don't measure production engineers by their ability to write code on paper, we give them compilers.

[Pruthvi @pvibhat]

@OmedVibeCodes Now use that same custom prompt for fable itself.

[Pruthvi @pvibhat]

@niclas__4 @composio Lol there's benchmarks, real-world uses and independent verifications from multiple credible orgs. There's no assumptions being made, that's just cope.

[Pruthvi @pvibhat]

@anjiyuci @kimmonismus Disagree. This is only true for US companies. The rest of the world (Europe mainly) still sees US based hyperscalers, with there data retention policies as a regulatory risk. This move cements that divide further. Anyone wanting sovereign AI is now at odds with the US model

[Pruthvi @pvibhat]

@thsottiaux I do other stuff while listening to GPT so the TTS for read aloud needs to be updated with the new better voices. Also, please make read aloud an actual background process so I can close the screen while listening

[Pruthvi @pvibhat]

@haider1 Your point is good but self hosting?? Bro which individual is self hosting at 2TB VRAM 😭

[Pruthvi @pvibhat]

@uoislame @ExistentialEnso Lol. AI literally makes code monkeys useless. Every tool has pitfalls, it's always been ROI.

[Pruthvi @pvibhat]

@iruletheworldmo I'm piling on but it's nice that you're open to changing your mind. Honestly it's kind of interesting how all of us never suspect American labs nearly as hard for benchmaxxing. Have you used k3 extensively yet?

[Pruthvi @pvibhat]

@cyrilXBT You're literally like 1 year behind with this... This is the kind of thing you'd do with Gpt4.1

[Pruthvi @pvibhat]

@SourabhGurwani Ask a subjective question... Get a subjective answer.

[Pruthvi @pvibhat]

@zoomerbread @CupasTopNigga @FakeSungate @AnCapShow Fair, but there's several reasons. I'm saying that attributing "Modern Software quality has gone down" to " they use LLMs" to " that's why LLMs = bad quality" is a dumb argument. Corporate incentives, ecosystem complexity and other reasons are also imp

[Pruthvi @pvibhat]

@CupasTopNigga @FakeSungate @AnCapShow Besides "I know the debugging process would be hell" and "I'm not a coder" do not go together either. Just say you don't like it for your own use case rather than LARPing as a technical person

[Pruthvi @pvibhat]

@CupasTopNigga @FakeSungate @AnCapShow "Not a coder" "Modern tech has gone down in quality". How can you talk about code quality without having worked with software?

[Pruthvi @pvibhat]

@Zed_Oh_Six @woke8yearold Isn't the reasonable argument then about where they're built rather than if they're built

[Pruthvi @pvibhat]

@GlenWilsonIA @Vathreon @thsottiaux I definitely agree with you, I'm just disagreeing that "these models aren't intelligent; patterns not reasoning" is obsolete. Reasoning even in humans is just patterns, what matters is the context and constraints given. Fable genuinely feels like a coworker when instructed right

[Pruthvi @pvibhat]

@Suchintan Incredible. What are the tradeoffs?