Хиена

Software engineer, AI enthusiast

[Хиена @_spookey]

@atuntable @bindureddy I have a theory that it's already happening in a way. You see, all data used to train models ultimately originates in the human brain. All text, voice, even the imagery is produced by brains. The least path of resistance for the algos to model it is to build a brain-like entity

[Хиена @_spookey]

@atuntable @bindureddy We need to consider the overall cost of employing a brain. Then compare it to a machine doing the equivalent intellectual labor - which nowadays is a single PC for about $2K paid once and running the latest Chinese model. Same speed for math, coding, etc. But it never gets tired.

[Хиена @_spookey]

@rohanpaul_ai You might need to go as far as to basically build a complete biological cell to obtain that 100% compliant artificial neuron, so next to impossible. And what if you replace 50%, 90% of neurons? Then there's the question if what you produced isn't just a p-zombie.

[Хиена @_spookey]

@mrandersoninneo @bindureddy There is some space for improvement, it's why quantization still works. But nature seems to be going the opposite direction. It's easier for evolution to duplicate alleles and carry around junk DNA than to compress and optimize it.

[Хиена @_spookey]

@lost_in_tech With FreeCAD and Python, combined with modern AI codegen, you get a really powerful tool for free. Stuff like optimizing parts dimensions to make an assembly move in a given path; or the same with FEA to maximize rigidity.. and that's just what I did for fun.

[Хиена @_spookey]

@bindureddy What do you expect when there are basically NONE open US models.

[Хиена @_spookey]

@JustinLin610 Thank you!

[Хиена @_spookey]

@pascal_bornet And this costs in the five or six digits, right

[Хиена @_spookey]

@huybery But models aren't code, they are large weight matrices. No model writes those and they are only made with very expensive training.

[Хиена @_spookey]

@Alibaba_Qwen wtf, only mac?

[Хиена @_spookey]

@IlirAliu_ Can't step over thresholds.

[Хиена @_spookey]

@AmericaPartyX Why does Elon appear on everything? Starts to look like cult propaganda.

[Хиена @_spookey]

@deedydas R1 did not have vision.

[Хиена @_spookey]

@huybery Longer then 8k token outputs. I'm hoping for it.

[Хиена @_spookey]

@humanfellow0 @JustinLin610 That doesn't mean it will work on the whole 10M. It's probably the same deal after >200K.

[Хиена @_spookey]

@bindureddy First thing to learn is how to use version control. Then you can undo/redo until it works.

[Хиена @_spookey]

@petergyang The same applies to game engines unless you vibe code the whole thing from scratch.

[Хиена @_spookey]

@JustinLin610 is it time?

[Хиена @_spookey]

@sippicup2 @pud But I don't want to solve basic bugs, or any. I want thoughts materialized without friction. Of course I'm lazy, that's the point of, you know, technology.