Peter

[Peter @kasselvania]

@michaelgofman @BenjaminBadejo I have literally seen the EXACT OPPOSITE, they are EXCEPTIONALLY careful, and are being detail oriented to authority and the claims there-in.

[Peter @kasselvania]

@BenjaminBadejo They have also claimed of “nerfing” because of context window decrease, token over usage (which was already called out by tibo) and how it is a “garbage model.” Then, X seems to be AWASH with this shit… on the day grok released a model… it’s all icky feeling.

[Peter @kasselvania]

@DaveShapi If I were to use a more blunt way to say it, it’s clear Anthropic is high of the smell of their own farts

[Peter @kasselvania]

@DaveShapi Honestly, I appreciate that folks think critically about it. It’s important. But the self aggrandizement has bled into outright pretentiousness verging on outright contempt for the user base. It’s a bummer.

[Peter @kasselvania]

@kimmonismus I read this as "we are about to get all the people at work showing up on Monday, had negotiated a bunch of new contracts in prep for it, and need to make sure that we can handle the load."

[Peter @kasselvania]

@thsottiaux Wink if we should turn on Fast mode and just not be concerned!

[Peter @kasselvania]

@thegenioo Of one of the most powerful models of all time… like, not even Luna or Tera…

[Peter @kasselvania]

@thegenioo How much inference are you expecting on $20…?

[Peter @kasselvania]

@kasselvania on Plus account? $20?

[Peter @kasselvania]

I have some serious hypotheses and ideas about this whole “codex eating tokens” thing.,. And it boils down to “users error in hygiene of model errors.”

[Peter @kasselvania]

@quantumaidev @notiyda Not really how this works… we all get the same limits. If tokens are getting blown, the model is most likely working harder. That often means a lack of consistency or clarity. Sol doesn’t want to assume or lie or make mistakes. Older models would. That specificity costs tokens.

[Peter @kasselvania]

@quantumaidev @notiyda 🤷🏻‍♂️ I’ve been building a strictly governed runtime/OS for a project, and it has run literally every minute of this weekend on xhigh sol and I have used 65% of weekly usage or something. Friday 11pm I started, right now, I’m at 35%

[Peter @kasselvania]

@quantumaidev @notiyda That doesn’t mean anything… if the authority, documentation and planning isn’t clear, it’s going to be spaghetti eventually, no matter who codes it

[Peter @kasselvania]

@ann_nnng Depends on what you mean. I had fable choke on multiple governance hiccups, while GPT called out its bullshit. “Smart” literally means nothing objectively and nobody has an objective restore proof either.

[Peter @kasselvania]

@quantumaidev @notiyda Means your code is messy. The more it’s gotta work to figure out your code, the more it burns.

[Peter @kasselvania]

@notiyda This was my experience. 9 hour, 30 min authority alignment pass. Huge. 256 files, three different phases of life for the product, so quite a few contradictions. Authority passes constantly. Perfect work, 22% weekly usage.

[Peter @kasselvania]

@thegenioo I literally don’t believe or understand this. I’ve been using xhigh Sol for the last 72 hours, and I’ve used only 70% of my weekly limit

[Peter @kasselvania]

@thsottiaux @RileyRalmuto Man, I ain’t complaining! Resets are resets!!!!

[Peter @kasselvania]

@jumperz Like, for REAL, most people assume that their fucking app needs TOP TIER INTELLIGENCE FOR ALL CODE, when high models are best for true deep planning, slicing, runtime or other complex meta tasks… not code.

[Peter @kasselvania]

@jumperz Terra on medium is damn near 5.5 Xhigh in terms of capabilities. If you’re on Terra High, you’re already matching, if not exceeding 5.5 Xhigh. All the rest is literally extra capabilities. Sometimes not even best for main coding.