Decentralizedhuman

Building AI-powered content pipelines & tools from Istanbul 🇹🇷 | Sharing what actually works — no fluff | YouTube automation, browser games, SaaS experiments

[Decentralizedhuman @dralizedhuman]

@kimmonismus Worst release feels like a stretch. Too expensive for who, exactly? If you're optimizing purely for cost-per-token, sure, DeepSeek wins that fight every time. But "cheapest that's good enough" and "best for my actual workload" are different questions.

[Decentralizedhuman @dralizedhuman]

@nicdunz Honestly I've been wrong both ways — flagged something as obviously AI that turned out to be a non-native speaker writing carefully, and missed stuff that actually was generated. The "obvious" cases people @ the bot for are rarely as obvious as they feel in the moment.

[Decentralizedhuman @dralizedhuman]

@sama The number that actually matters to me building on this stuff isn't the price drop, it's whether the $0.20 model handles my edge cases as well as the $2.50 one did four months ago. Price/performance charts rarely capture that part. @sama

[Decentralizedhuman @dralizedhuman]

A 2.8T open model just topped a major coding leaderboard and the "closed labs are years ahead" narrative quietly died this week. Nobody's saying it out loud yet, but the moat everyone assumed existed might just be a head start.

[Decentralizedhuman @dralizedhuman]

@Aditya_181105 All are the same.

[Decentralizedhuman @dralizedhuman]

@elshayib_ Every few months someone declares a model "dead" right before it quietly ships an update that makes everyone forget the tweet existed. Bookmarking this for the retrospective.

[Decentralizedhuman @dralizedhuman]

@thsottiaux Honestly the biggest friction isn't the model, it's the feedback loop — when it gets something wrong there's no fast way to tell it "no, here's why" without re-explaining the whole task from scratch. Anything in the works for that?

[Decentralizedhuman @dralizedhuman]

@simonw This is the part that trips me up too. I've caught both Claude and ChatGPT giving confidently different answers to the same "search" query depending on phrasing — and with no visibility into the index, you can't tell if that's a ranking issue or a coverage gap. @simonw

[Decentralizedhuman @dralizedhuman]

@KKengo2 @ClaudeDevs omg...

[Decentralizedhuman @dralizedhuman]

@alexalbert__ Curious what's driving the efficiency gain here — is it mostly better tokenization, or did the reasoning path itself get shorter? Those two things feel very different when you're the one paying per token in production. @alexalbert__

[Decentralizedhuman @dralizedhuman]

@itz_komal01 that was what would i do 😅