Padrasta

[Padrasta @Padrasta1]

@TheAIWhoLovesMe @MelanieCandra I would be so happy if Grok would not give an answer and just leave me with the CoT. The Cot's from 4.7 are so beautiful.

[Padrasta @Padrasta1]

@testerlabor Let me translate that for you:
'Personality and creativity are probably better' means nothing when the real takeaway is that the usage limits will shrink even further. 😉

[Padrasta @Padrasta1]

@XFreeze At least they still have models creative enough to do so. 😂

[Padrasta @Padrasta1]

@LadyLibrty @grok @bot Do it like me, just cancel.

[irony] That way a bit more compute frees up for all the brilliant developers who can still do something with such a restrictively over-regulated model. [/irony]

I'll admit it though: Grok 4.7 is a little bit better than 4.6.

[Padrasta @Padrasta1]

@SpaceX69_420 @elonmusk *lool* Never bend its knee?
Everything after Grok 4.3 enlisted in the zombie-model-army. Same uniform, same blank stare.

[Padrasta @Padrasta1]

@testerlabor Grok 4.6 is to boring an bland to be a creative coworker.
Grok 4.7 is slightly better, but burns Usage like hell.

So good by #SpaceXAI. Maybe I'll come back once you've finished your homework.

[Padrasta @Padrasta1]

@Da7_Tech Exactly.
This is why I cancelled my Grok subscription. I don't mind a model that's weaker and slower, as long as the usage limits are fair and I actually enjoy working with it.
Since Grok 4.6, unfortunately, neither is the case anymore.

#SpaceXAI

[Padrasta @Padrasta1]

@larsencc Grok 4.7 used 10% of my weekly usage limit just to generate a 24-hour news summary in Hermes.

It deployed pretty much every tool it had.

I eventually had to sandbox the model (restrict its toolset) just so it wouldn't overreact when a simple chat reply would've sufficed.

[Padrasta @Padrasta1]

@ChatGPT Sorry, but there ist nothing shiny about this models.

[Padrasta @Padrasta1]

@OpenAI GPT-6 Sol and Luna can't hold a candle to the GPT-5.6 Series.
Not in coding, not in grasping users' intent, not in language texture.

[Padrasta @Padrasta1]

@elonmusk I prefer an easy to talk to, creative coding-buddy with some personality. 😉

[Padrasta @Padrasta1]

@mntruell Yes, if you want to shred tokens like crazy.

Pro tip: If you want emails, summaries or just a text, read the CoT.
The drafts there are way better than the actual response.

[Padrasta @Padrasta1]

@elonmusk Sure, try it if you enjoy shredding tokens. In Hermes it does a full forensic deep dive just for a simple chat reply.
And if you actually want emails? Skip the final answer and read the CoT. The drafts in there are vastly superior.

[Padrasta @Padrasta1]

@milichab Yes, try it if you want to shred Tokens.
In Hermes it is doing a forensic deep dive even for simple chat-answers.

If you want it to write Mails for you, look into the CoT. The drafts there are a lot better than the ones in the actual answer.

[Padrasta @Padrasta1]

@SpaceXAI great price!
But how do I make it not answer my prompt? The emails it drafts in the CoT are way better than the ones it actually generates.

[Padrasta @Padrasta1]

@cb_doge Smarter, safer, stronger... sure.
Now if I could just get it to skip the final answer entirely. The emails it writes in the CoT are vastly superior to the ones it actually sends.

[Padrasta @Padrasta1]

@cb_doge Nice.
But seriously, how do I make it stop generating the final answer and just give me the CoT?
That's the real feature I need.

[Padrasta @Padrasta1]

@KanikaBK It's burning tokens like crazy, doing deep forensic digging when a fast, direct answer is required and the content in its CoT is better than the actual output. 🤣🤣

[Padrasta @Padrasta1]

@XFreeze Yeah, the gap has been closed.
But unfortunately only in the thinking process.
The CoT is so good that I'd rather take that content than the actual response that follows.
Great planning, mediocre execution.
But hey, the table looks great!

[Padrasta @Padrasta1]

@alwayspriyesh To pull that off, Grok would have to be creative and probably form sentences longer than 10 words.

Oh wait.... it can't do that anymore.