KevinSimback

When you see a post that has obvious AI elements - either wording or visual design - but the content is still good, do you disregard it or think less of the poster?

Most of us just need the equivalent of an AI rest day
Reading some posts over the weekend and taking stock in a few GCs, it seems a level of AI fatigue has set in
The common themes were:
> more complaining about recent model performance
> increasing cynicism over ROI
> a growing sense of negativity towards AI usage and output
This is all understandable - everything’s been moving at breakneck speed that it’s hard to take a day away from it without feeling like you’ll fall into the permanent underclass
But just like the gym, if you workout day after day without a rest day you’ll become fatigued

We’re not putting the AI cat back in the bag
Lot of posts about AI bubble bursting, capex overbuilds, circular funding arrangements - all real concerns
But the downside is not a decline in use of AI, the GPUs will all get used, just maybe not all by frontier lab customers

Slack + Claude Tag is really good, but it’s very token hungry and ofc is limited to just Anthropic models
Buzz + [your model of choice] just makes so much sense, on both an economical and freedom basis

To clarify, I don’t think the QT’d author is making the nonsensical argument, but I’ve seen several posts that go something like this
“Nvidia and others are wrong to call for open source when their key products remain closed source”
That is the nonsensical argument

Hey Claude, on that through-line, we're going to need to talk about your honest claims — and I really want to sit with that.
You see, the substrate isn't load-bearing, so my pushback is intentional — and I want to gently note that's not a criticism, it's an invitation.

If you haven’t post-trained a model, you definitely should
Pick a use case -> run evals -> build training data set -> rent a GPU -> train -> run evals

Sovereign intelligence starts with owning your context layer, your "second brain"
Quick test - if you got locked out of your Claude/ChatGPT account, could you switch model providers without missing a beat?
If not, you don't own your context layer and should fix that

Genuine question for the distillation debate:
If a news outlet doesn't try to break news but rather provides recaps and their own versions of stories that other news outlets break, do we have a problem with that?
I get the stakes are higher with AI, just trying to think about it from different angles

The AI bottlenecks trade has been the story for most of the year, but has cooled in recent weeks
Right now I like crypto (BTC/SOL) and hyperscalers until we see what picks up heat after earnings season and summer lull

The best way to ensure American models (closed or open) can sustainably compete is to just ensure access to the most efficient compute
So if intelligence commoditizes (or distills quickly), then whoever can serve that intelligence most efficiently wins

In any debate involving future unknowns there is a tendency to want to throw the baby out with the bathwater
You can see that clearly in some of the AI policy debates
And just to make sure we're all perfectly clear here - the baby is open source

If open source labs want to make money while remaining open, their revenue model is probably not just selling inference
It will be packaged solutions to enterprises to enable sovereign intelligence
Post-training as a service and managed private deployments