The Clinamenic Co-Living Diary

[catalog]

Sufi are you pro gen AI?

 

[william_kent]

i made this https://cotds.dissensus.com/ using that technique

there is too much intention for it to be automatic writing/coding - you are requesting things which the ai carries out in its mysterious way,
so if the ai is like your subconscious code brain you're still steering.
and you have to make your requests ai compatible: in careful bitesize chunks with all ambiguity removed or the ai gets lazy and comes up with shit or gets confused, so ultimately its a form of submission to the llm when you are dismantling your concept into ai format

but it is super useful for coding 🤩🤩🤩

I went to an "inaugural lecture" where the prof produced a powerpoint show showing the reliability of LLMs for coding reliability

Ollama won out at 10% accuracy

I spend more time arguing with chatGTP than it would take for me to just knock up some code

#old man complains
#shouting at the sky

 

[catalog]

Bill what are your thoughts? Are you "wait and see"?

 

[sufi]

Sufi are you pro gen AI?

what even is that?

 

[catalog]

Are you positive about generative AI?

 

[sufi]

Are you positive about generative AI?

no idea

 

[william_kent]

Bill what are your thoughts? Are you "wait and see"?

I was cited in papers ( well, an abstract in "cute studies" - I got a weekend at a conference out of it ) in 2018 for my work with generative AI so yeah, I was an early adopter, and I did start a thread I think about this stuff maybe?

 

[catalog]

Yeah but I'm conscious you don't seem to be getting on bandwagon? And sufi neither

 

[william_kent]

Yeah but I'm conscious you don't seem to be getting on bandwagon? And sufi neither

it's the new "blockchain"

I may be a bit jaded when it comes to empty air hype

 

[sufi]

I went to an "inaugural lecture" where the prof produced a powerpoint show showing the reliability of LLMs for coding reliability

Ollama won out at 10% accuracy

I spend more time arguing with chatGTP than it would take for me to just knock up some code

#old man complains
#shouting at the sky

i think if i was a better coder and better project manager and more patient and had a paid account, i could get proper nice code out of the llms

but they also make basic coding accessible and fun for absolute non-coders which is awesome but quality is gonna be low

 

[william_kent]

I work with medics who are trying to rid the world of disease and one of my colleagues had a meeting with some guys from Microsoft and they had free azure credits to dish out and they asked 'what are you doing to cure cancer" and my colleagues showed them some fantastic tools but the MS guys cut them off mid presentation and said "but our bosses at microsoft only want to know about AI! we don't care about curing cancer! just show us a project using AI and we will give you free money!"

so I'm sort of cynical

 

[sufi]

Sufi are you pro gen AI?

when you put it like that it sounds like "do you support the genocidal elimination of the Palestinians by highest specification military technology"

 

[william_kent]

i think if i was a better coder and better project manager and more patient and had a paid account, i could get proper nice code out of the llms

but they also make basic coding accessible and fun for absolute non-coders which is awesome but quality is gonna be low

the code looks correct and it will compile but it's just based on a load of questionable stack overflow posts and I spend more time debugging then it would have taken to just write it myself

but then I'm a bit cynical

it's like having your own useless intern

 

[catalog]

OK thank you lads I'm sorry for discourteousness

 

[Clinamenic]

i made this https://cotds.dissensus.com/ using that technique

there is too much intention for it to be automatic writing/coding - you are requesting things which the ai carries out in its mysterious way,
so if the ai is like your subconscious code brain you're still steering.
and you have to make your requests ai compatible: in careful bitesize chunks with all ambiguity removed or the ai gets lazy and comes up with shit or gets confused, so ultimately its a form of submission to the llm when you are dismantling your concept into ai format

but it is super useful for coding 🤩🤩🤩

Cursor has a "YOLO Mode" setting which essentially just lets the coding assistant make a bunch of subsequent tool calls without user approval (you can set basic parameters, e.g. allowlist for terminal commands and whatnot), so to a certain extent you can have the AI code stuff without user guidance, but there are still limitations.

 

[version]

Between me and the machine there is no estrangement. I am the machine.

 

[Clinamenic]

Exactly. A friend of mine used the term "cognitive prosthesis" which I think makes all this a bit more interesting to Dissensus

 

[catalog]

Yeah it's good Stan, show us some of the fonts it's made?

 

[Clinamenic]

I'm still making the core functionality, and so far I've just dummy fonts with it, to test vector graphic creation. But this already makes it so much easier to make fonts, compared to how I've made my four existing fonts.