xlbuttplug2

So far the next tier has only demonstrated that they can catch up to, but not necessarily leapfrog, what the top tier has put out publicly.

I suspect the top labs will come up with a business model that doesn't involve handing out their secret sauce for everyone else to reverse engineer. Perhaps restricting their top models to select high paying government/enterprise contracts. Or maybe a bespoke "describe the problem and we'll solve it for you" type service.

Eventually the top labs are going to collude and simply not release their best models to the public (if they aren't doing that already).

I presume post training is significantly easier than the distillation/training the top Chinese labs are doing.

I wonder if, similar to the American labs, they'll become stingy with their weights once they start getting immediately undercut by a wave of slightly better derived models.

Well, are you planning to do something with this information or are you just claiming to be self aware? :)

Well this got buried quick..

I'm lucky enough to not notice my tinnitus except when I try to do nothing.

I think you're equating doing nothing with doing nothing productive. You're usually still ruminating or trying to distract yourself from the source of the pressure.

I wonder how long it'll be before AI labs find a revenue stream that doesn't involve renting out their models, and stop letting us ride on their coattails.

They'd tease us with solutions to hard problems, with code that is orders of intelligence higher than any human or public model can grok.

That'd turn all the whining to begging real quick.

I'm guessing the next generation of US frontier models will be heavily anti-distillation at the cost of user experience (significant rate limiting, more flat out refusals, hidden thinking, etc).

And as long as they maintain a significant advantage in capability, we will continue to kiss the ring.

May not be very effective if so.

I'm assuming finding vulnerabilities in open source projects is the hard part and what you need the frontier models for. Writing an exploit given a vulnerability can probably be delegated to less scrupulous models.

My hunch is it's a combination of

* coming in with a bias of not wanting it to work

* having too high of an expectation

* giving up too early

* not trying SOTA models

* not taking the effort to communicate intuitive or painfully obvious things

But perhaps it is too dumb to solve the type of problems you guys are working on and no amount of cajoling will help. All I know is "it works for me."

In my case, the volume of code is roughly the same. I'm not using the efficiency towards pumping out more code, just using it to be AFK more.

I spend enough time iterating and refining to the point I'm comfortable taking ownership of the outputted code. Perhaps hypocritically, I do mald when people upload code for review that they clearly haven't taken the effort to read through critically.

More like a day's worth of coding condensed into half an hour. Time to review/test is mostly unchanged.

Day to day is mainly minor feature additions into a stable product so not a huge amount of code churn.