Jon Claassens
@JonathanClaass4
Followers
183
Following
6K
Media
148
Statuses
6K
Indie games, 3D graphics, robotics, business software
San Francisco, CA
Joined February 2020
Windows edition of our big 3D editor/CAD coming along, but some way behind Mac for now
0
0
2
To my point, 10k lines a day is 30 days away from a rewrite of SQLite, a year away from MySQL. Armies wrote those systems. Literal armies.
0
0
0
If you're generating this much code in a day, the code is "extremely likely" given the lack of need to check it and must be pure boilerplate. End of story.
I met a founder today who said he writes 10,000 lines of code a day now thanks to AI. This is probably the limit case. He's a hotshot programmer, he knows AI tools very well, and he's talking about a 12 hour day. But he's not naive. This is not 10,000 lines of bug-filled crap.
1
0
1
High power MacOS/Windows version of Konjur3D coming along. Metal native renderer here. The Python scriptable interface is being carefully crafted to be intuitive
0
0
0
As prophesied, a chosen one will unleash the singular mind, the AGI of AIs. The great tech companies Meta, Google, and Microsoft seek his greatness with gifts of gold and glory. But who is he, and will he bring a new dawn or a bleak future of trash and even worse TV.
0
0
0
If AI does actually reduce code cost, that'll impair the intangible asset value of newer businesses that don't rely on revenue to model asset value. It'll also reduce the 'moat' or barrier to entry cost to attack their market, which will reduce their revenue outlooks or DCFs.
0
0
1
You're not doing anything hard enough unless AI is hallucinating garbage on your prompts and Stack Overflow fringe dwellers are telling you to rtfm or kys. .
0
0
0
It's their beautiful PowerPoints and constant alignment.
guy who thinks you hire mckinsey for their insights.
0
0
0
Talking to Windows developer: I'm making a table. Them: Oh, with Xaml?.No, #include <commctrl.h>
0
0
1
You always have the good guys and the bad guys and everything in between.
0
0
1
Something should navigate the prompt like a boat on a sea of stats.
0
0
1
This leads us back to a reinforcement learning paradigm, but along the dimensions of correct context, correct behavior and action inhibition (the latter being simply I dont know).
1
0
1
Further, I think another side is missing. The "filter" could be a validation structure (and many companies do that with output validation classifiers), however, there is a dimension beyond mere validation necessary. LLMs fail when the context is inappropriate, that's the key.
1
0
0
LLMs are a noisy n-gram channel shaped by prior "prompts." The joint probabilities are hidden under a pile of weights. However, filters on the output can be constructed to reduce noise. I think this is what the agentic architecture entails. .
1
0
1