Thoughts, Reflections, Rumour
my trip through chatGPT (cgpt)
it was okay. certainly vaguely capable. bright future. has a similar drag as video editing.
it says nice things about my language system design, which seems too complicated for it at the moment.
it will perhaps need training on my language to really get into it (I have just now discovered the docs on fine tuning)
I should just do my work, I think.
immediately
far sooner than mere human.
it doesn't muse about code so well
keen to discard facts
good for immediately checking spec == code
by asking it to change anything it wants
it may be best at writing code that is already written. that seems a pretty full function - look around for improvements.
but it has a style and structure then, whereas giving it wider scope wanders back to some muddy normal.
kicking off the writing of voicing() seems to be my goal
how exciting. I could be into it already. some apparatus of multitudinous movementisms for the mind <-> world (the world of A+C).
perhaps it needs to be molecular, a few A walk each other seeing what to do to the other.
to split the 'thinking of you'
however
it presents such an immediacy, tooling for which is not invented.
reminiscent of using a video editor made by a primate, so much carpentry missing.
it has to be coaxed into remembering specifics by using them for something.
specifics (eg subroutine argumentation) can wander inexplicably.
writing good language never fails. surfing into efficiency. getting a grip on the slope.
not writing good seems to annoy it and beget bad replies.
I resolve to wander back to not using cgpt for the next week or until I think of something.
I would continue with regular immediately checking spec == code if I had more automating on the ghosts I am typing, which sounds reasonable. maybe in a good week!
Ultimately the universal should not leave.
improvements
some apparatus of multitudinous movementisms
needs strapping on top of this api
variations that are then asked to be modified for some purpose, favour code with efficient evolution
variations to check a whole lot of aspects we know are there for each moment (relieving the origin moment?) eg "could this be more general", "are these arguments correct", "are you still writing typescript"
variations just feeding it differently. "chew on this" style. try various chunks of whatever can be found in the way of spec or code by text matching, provides entropy and maybe sparks.
varying levels of repetition of the core of the language might help:
strap into the language and things we are working on
which is I suppose a purpose of the A as we have been chatting about.
see stylehouse.git
defining the defined more definitely is needed.
ie it wanders back to some muddy normal. peddling stuff.
might be related to the I don't have access to any previous conversations limit
perhaps there is an expensive model of the code being developed needed
it might need to converse with some model I run longer term.
can they bifurcate? be run together, as if not trained separately? ie regrouping chunks of neurons, on the phone
building the internet between some crazy realm of ideas that the main pile can't fully know and putting the two conceptual spots in communication as such.
>1 hour
unsure what the limit is. page reload required, no conversation resume.
maybe it always gets babylon once yay long and winding? is it a constrictor? made of knots that get tighter, get annealed?
bring the nebulous
uses
no one gives you breaks
as an artist. virtual boss for freelance workers.
bots
interprets overall meaning quite well
to get the cue, ie someone links youtube in irc
or on youtube, when someone asks what song is playing when.
secretary of syslog
should I care about Mar 5 11:56:06 sa kernel: [ 2.525542] psmouse serio2: synaptics: Your touchpad (PNP: TOS1151 PNP0f03) says it can support a different bus. If i2c-hid and hid-rmi are not used, you might want to try setting psmouse.synaptics_intertouch to 1 and report this to linux-input@vger.kernel.org.
secretary of scanlan
about lumps detected, camera's ideas wrt life.
can we haunt away the security situation.
quotations
on voicing()
This suggests that the throat function has a crucial role in mediating between the mind and the external world, allowing for the interpretation and expression of ideas and concepts.
how ethnographic
(I have just now discovered the model’s maximum limit ... 4096 tokens- then that GPT-4 is 32k tokens, and there's also Reformer, a Transformer model designed to handle context windows of up to 1 million words)
reason flies in and discovers itself everywhere
Assembling reason and discovering itself may involve navigating through complex systems and making sense of a vast amount of information.
multi hazard multi angle, while remembering who you are while playing with toys.
It may also require creativity and the ability to generate new ideas and insights.
insight seems to be at a premium.yay
writing good language never fails, it's only after that first blissful moment that anything goes awry (unless it's already awry from fragment-stuffing), perhaps bad writing scares it.
you can begin to reveal how to do things quite freely:
lets show a pdf in Svelte by shelling out to `convert` to make png of the page.
import { execSync } from 'child_process';
function pdfToPng(pdfPath, pageNum) {
const pngPath = `${pdfPath}-page-${pageNum}.png`;
const command = `convert -density 150 "${pdfPath}[${pageNum - 1}]" "${pngPath}"`;
execSync(command);
return pngPath;
}
<img src={pdfToPng('path/to/myfile.pdf', 2)} alt="Page 2 of my PDF file" />
It goes!
Some error on the client about Cannot access "child_process.execSync" in client code,
which we propose to work around by doing it all on the client with:const pdf = await pdfjsLib.getDocument(pdfPath).promise;const page = await pdf.getPage(pageNum);
Thus I put pdfToPng() into a +page.server.ts of letz 276e295a27c238b
A win.
People have said "expanded my ambitions"
So
chatgpt would struggle to remember all the facts about the programming language thing we were developing. keen for more tokens.
I am keen to move on from using the webapp to automating dialectics, eg for:
- scrubbing confusions as they emerge
- templating phases of talks, ie games
- gisting what has happened in the convo so far if human attention is coming in
- using lots of writing (spec|doc) I have around
- rolling the lingo around a bit to find optimal expressions
- finding efficient (min cut) ways to divide the mesh of information that defines the thing we are developing so parts of it can be intensely illuminated and worked on (ie weld).
And if you made it to my blog thank you for making it to my blog.
No comments:
Post a Comment