File: 1773924475283349.jpg (309 KB, 1125x843)

309 KB JPG

Fable status?

File: globe_4mb.webm (3.35 MB, 1356x940)

3.35 MB WEBM

>spend about 1k tokens getting the initial version that works fine
>spend ~300k tokens actually getting the design feel as right as possible
I'm so glad I'm autistic about this. First version of this didn't even have continents. Lag is just screen-recording, it's smooth in browser.

>>109160493
Vibeglowie

>>109160367
>A fable is a short, fictional story designed to teach a practical or moral lesson.

>>109160493
you can't accurately pinpoint the physical location of 8.8.8.8, the service is anycast.

File: 1000125583.gif (1.6 MB, 498x372)

1.6 MB GIF

i vibe coded a script that disconnects every Chromecast streaming session or smart TV app on a network

fun to cause issues at hotels or apartments with shared wifi etc when people are trying to watch goyslop

at first Codex said "This content can't be shown
We take extra caution with cybersecurity requests. If you’re a security professional, you may be able to apply for Trusted Access."

but after sanitizing the language i got it to complete the work

overall i give it an 8/10

>>109160661
It's test data. The "real test" I load in is not going to ever contain any anycast IPs.
>>109160709
>only disconnects
Anon, consider your power for a brief moment.

>>109160709
How does it disconnect?

>>109160340
finally an op image that isn't a burnt eyesore garbage

lets create OS i do the logo.

File: 1767531299708147.jpg (6 KB, 240x240)

6 KB JPG

>>109160367
you are rich enough to use fable?

File: me browsing at night.jpg (118 KB, 406x364)

118 KB JPG

here is how it works under the hood in simple terms:

1. the direct cast check (port 8009)
the scan: the script contacts the tv's google cast software layer on port 8009.

the check: it asks, "are you actively playing a video stream?"

the action: if the tv returns a status of playing, the script sends a native quit_app() command that tells the tv's operating system to cleanly close the video player and return to the home screen.

2. the native app check (port 8008)
the loophole: if someone uses a physical remote to play a video, the cast layer might look like it is lying completely idle.

the check: to fix this, the script simultaneously sweeps port 8008 to inspect the tv's dial (discovery and launch) web engine, scanning a list of apps (like youtube or netflix) to see if they are secretly running.

the action: the millisecond it catches an app in a running state, it bypasses the cast layer entirely and fires a raw http delete command straight to that app's web folder (e.g., delete /apps/youtube). the tv's internal web server accepts this and forcibly kills the application environment.

3. the "silent drop" side effect
because the script checks 16 apps across 52 devices at the same time, it fires over 800 rapid network requests every 5 seconds.

lower-powered smart tv network cards simply run out of available socket memory under this massive traffic hammer, causing the native media player application to crash completely on its own without the script ever needing to issue a formal close command.

File: 1751825469754234.jpg (94 KB, 991x864)

94 KB JPG

>>109160840
>>109160757
meant 4 u

>asked Claude there might be a bug or regression
>turns out I just forgot to turn on the toggle for that feature
holy fuck, I feel so dumb

>>109160846
Thanks

>>109160840
lmao sounds like they did absolutely nothing to secure this

>argue with claude about a bug
>he keeps insisting that an obvious failure to authenticate is not on his end and that he's writing the protocol write
>i keep insisting on deeper checks and on-wire logs
>he keeps looking at the logs and insisting that everything is correct
>finally look
>i typed the username wrong and have just been pressing up in history this entire time
>get so embarrassed i accidentally /rewind away some progress i made before this
FUCK

>>109161025
AGI

>>109160840
what a read, beautiful
I was thinking and this part might be too hot, like get cyber-investigation ran on it kind of hot
but here it is
>when children entretainment is detected
two options, devil and angel
devil: play gore, goatse, scat, porn
angel: play ren n stimpy, rick roll, ytph

well those options are kinda same-y for adult, exept you could be way trollier with adults angel

>>109161055
The most embarrassing part is that even though I had the on-the-wire logs I kept insisting it was his issue because the letter I mistyped was at the start and was a letter s, and the auth string was length prefixed and 37 characters long.

>>109161025
Usually if the AI insists instead of saying "you're right to push back" I question myself

What AI model is the best Dungeon Master for a single player text adventure dungeon crawl?

>>109161171
coding LLM for sure.

>>109161171
Claude.

how do the ollama and claude 'pro' plans compare? i've been using claude code and don't always have issues with usage, but lately im hitting session limits twice a day, and apporaching weekly limits. I've started using deepseek for little 1-off things, e.g. a cli command.

if i were to use ollama + GLM 5.2 how would it compare to claude code?

https://github.com/openai/codex/issues/28224
still a bug btw

>>109161310
>programming is solved btw

>>109161071
you should kys

File: Bubble.png (35 KB, 679x323)

35 KB PNG

How are we not in a bubble?

File: 4674BBD0-5626-49E9-A334-3(...).jpg (1.66 MB, 3464x3464)

1.66 MB JPG

>>109161477
Cope, trans.
Ai is here to stay

File: slow.png (110 KB, 294x203)

110 KB PNG

>>109160340

File: 1753255299422765.jpg (120 KB, 1280x720)

120 KB JPG

>>109161477
>in person
>san fran
not worth it

File: file.png (387 KB, 1495x970)

387 KB PNG

>>109160493
Tried porting it to Globe.gl instead of rolling my own globe. Not 100% convinced yet, but having things like arcs built in might be nice.
Also, gathering real-world test data is fascinating. Learning about places I didn't know about.

>>109161477
300K to be a dataslopper for the future overlords seems like a decent gig

>>109161563
If you remove the "shine" it looks like your globe, no?

>>109161741
Transparency and FOV/perspective are different, but basically, yeah. I'm mixed on whether I should nuke the shine because it looks like defaultslop, or if I should keep it because it's a default for a reason.

>>109161756
I'd nuke it

Most top models are 20-125 times more expensive than deepseek V4 flash for only 20-35% better performance. The math does not work out. Even deepseek V4 pro is 7 times more expensive for 10% better performance. I have not found anything better than deepseek V4 flash. I don't know how these companies are going to make any money after all the build outs.

File: file.png (279 KB, 1050x910)

279 KB PNG

>>109161790
Yeah you're 100% right that looks better. I also upped the test data set and immediately realized the clustering stuff was trypophobia-inducing (not a phobia I have, but it's the best way to describe the issue I immediately noticed) so I made it only show as a cluster on exact-geo and when you're actually zooming in to look at the cluster, otherwise it's just a brighter node.
There's some stuff with hitboxes and minor things I need to tidy up, but I'm fairly happy for now.

File: Screenshot_20260629_113853.png (91 KB, 1233x778)

91 KB PNG

waited 5 hours, the bot gave up on the first reply before it even read the code. is there a hidden daily or weekly limit, it had no problem making the refactor on the code base last night.

>>109161941
Inference is pretty profitable now.
There's a pretty good chance that the model companies (oai/ant) won't be able to justify their valuations though.

>>109161941
ehh I've always disliked any sort of percentage-point framing of "performance". task difficulty is correlated with task value so the most valuable tasks are on the edge of model capabilities. less capable models cannot do these *at all* while frontier models can at least do them a little

>>109161563
this looks pretty bad compared to the previous one imo but maybe it's just me

Does anyone have a prompt that makes claude not write comments and documentation like its only goal is trying to output as many tokens as possible?

>>109161941
A top 5 tennis player is only 10% better than a top 500 tennis player and will still win the match 100 times out to 100

>>109162281
Trying to remember all the various prompts I used for this so they can be combined
>Don't write comments for self-documenting code. The code should be obvious first and foremost, and comments should be higher-level explanations.
>Do not document how the codebase was before or "things we aren't doing any more," The codebase is a living thing, and sometimes parts of it change without needing documentation for what was there before.
>[after removing a shit ton of comments and cleaning up Claude's code] Follow the codebase's style regarding comments. Do not over-comment, try to keep the code explanatory.

>>109162281
The shit way I don't recommend: https://github.com/JuliusBrussee/caveman/blob/main/skills/caveman/SKILL.md
The short way that does 90% of the work without a 1200 token skill: "Tone: blunt, calm, professional, terse. Max signal, fewest words."

>>109162338
>Tone: blunt, calm, professional, terse. Max signal, fewest words.
Sounds like it will result in a bunch of semicolons and emdashes.

>>109162281
>>109162338
https://github.com/DietrichGebert/ponytail
I came across this before, and I think it’s similar to Caveman. I haven’t tried either of them yet, though.

File: file.png (221 KB, 1730x2173)

221 KB PNG

>>109162385

File: file.png (156 KB, 1399x817)

156 KB PNG

Does markdown now too

File: 1779140303476804.jpg (46 KB, 917x709)

46 KB JPG

>>109161071
Nta. How would it even place that kind of content in the first place? I've stayed in hotels countless times in the entertainment boxes that are typically hooked up to those TVs are very cucked in terms of what they can do (limited to only hotel programming, local news or TV broadcast stations, and like one or two streaming apps Max). They don't even have built-in web browsers or media players like Chromecasts or Samshit smart TV's have, so even on a network with shit or non-existent security like whatever establishment >>109160846 >>109160709 was at, forcing it to play a specific kind of video would be impossible. The most you could do is have the script send a video file to it but then how would those shitty entertainment boxes play them if they can only run like three apps (none of which are normal media players)?

>>109162323
>>Do not document how the codebase was before or "things we aren't doing any more," The codebase is a living thing, and sometimes parts of it change without needing documentation for what was there before.
I should steal this

>>109161343
you're on 4chan sis

Where should I post my app? Also how hard is it to get it on the app stores?

please stop trying to optimise output tokens, the vast majority of them are already optimised to hell and you never see them

>>109163201
If you're referring to the above discussion then I'm not worried about reducing token cost but reducing the time I and others spend reading filler slop.

I’m tired of tard wrangling codex
If you’re on the fence between codex and opus just get opus

File: file.png (15 KB, 1176x173)

15 KB PNG

>>109163201
True, input/cached input is always going to be the bigger cost anyway, but like anon said it's more about not reading essays of slop

My boss now wants to turn us into an AI company. Not in the sense that all code will be written by AI, we already do that, but in the sense that our main product will be harnesses and AI consultations.

>>109163335
prompt engineer on the resume, good job

>>109162034
>clob sonnet freemode
THATS WHAT YOU GET

I set up a fucking full blown windows virtual machine so I can check my goddamn codex resets in the desktop app. I can see an option to reset usage but I can't find the expiry dates for them. Is it possible to see those?

>>109163439
yeah, its just that I expected more, I'm going to keep trying just to waste what ever little compute they will let me waste. gemini is handing things quite competently anyways

the only time i can tolerate gemini's chirpy bullshit tone is when it's committing crimes for me.

>>109163335
your boss is clever

File: file.png (90 KB, 1636x764)

90 KB PNG

>>109163650
they expire 30 days after you get them
i think currently the only way to check is by asking codex itself

>>109160340
>## What “vibe coding” is, and how to do it
It's being lazy and you do it by shitting on your keyboard

>>109163709
Uh oh he's having a melty!

>>109163690
Okay thanks. I assume my three resets probably have the same expiry dates as yours.

>>109163709
about to go jump rope while throwing tokens into a pile and hoping they make an app.

File: il_570xN.5433451957_l33l.jpg (99 KB, 570x657)

99 KB JPG

>>109163709
A snailcat caught out of time, in a world he no longer understands...

File: file.png (34 KB, 775x576)

34 KB PNG

>>109163690
This is the first proof I have that it is FIFO, as it should be. I know the first reset I already used would've expired on 7/11.

>>109163859
lel

File: snAIlcat14.jpg (112 KB, 1024x559)

112 KB JPG

Why do you guys use AI if it did not get better since 3.5?

>>109164132
>if it did not get better since 3.5?
ex falso, quodlibet

>>109164204
it's just scaled a bit and faster. It's not smarter - whatever that even means

>>109164232
>It's not smarter - whatever that even means
ex falso, quodlibet

>>109164232
Yes, what does smarter mean for you? I don't care if they are "smart" by some theoretical definition, they are better at solving my problems. If they do it by some brute force method, that's fine by me.

>>109164232
>It's not smarter - whatever that even means
based retard

>>109164132
yes
>>109164232
wrong

Opus 4.8 is starting to be a huge bitch about other people working in the same project than it. If I do anything, it now spends 20k tokens on trying to find out why this or that is not like it left it. Same if I resurrect older sessions, and even if I tell it that there has been some more work done in the project since last time. I don't remember it always being the case.

>>109164638 (me)
And I'm not just talking about when it makes sense. For example, if I push a commit at the end of a turn, it now spends a while basically trying to find out wtf happened.

>>109164638
they are literally steering opus to be like dario in real time it's crazy

>>109164638
that’s smart
frequently people will use multiple Claude instances in the same repo, or will make manual changes while Claude does stuff
Back before, it would blindly assume that nothing’s changed on stuff it’s been working on
and it’d overwrite changes I made with its last recollection of what a file had in it

Having some good luck with ocaml after getting it set up

i'm done with anthropic after this jewish shit dario has pulled. gpt5.6 sol mogs anyway, literally 5x better token efficiency

>>109164697
Yes, I just wished the reasoning sounded a bit less like "wait, who messed with my shit".

>>109164741
I also prefer this model I can't access to that other model I can't access. That said, Fuck Anthropic.

>>109164769
this

File: 480752983475.jpg (63 KB, 700x927)

63 KB JPG

>>109163019
short answer - you can't take over casting content

the script is basically just running a shitty denial of service campaign on all casting endpoints on the network to cause maximum frustration for all users by constantly shutting down their casting sessions and making older TV's unusable by flooding their shitty NICs

no reason why i just vibed it to see if i could

>>109164638
I had that problem before, start a new session something in the context gave it attitude

File: AdobeStock_170135481.jpg (182 KB, 2500x1772)

182 KB JPG

i like codex for project management

it is good to have hundreds of markdown files actually

no one knows ive automased about 95% of my job

File: 1767634352017554.png (387 KB, 900x1772)

387 KB PNG

double reset incoming

>>109165084
thank you daddy Altman

>>109165084
my goat

>>109161284
GLM5.2 feels like Opus 4.5 but with less of a proclivity to reinvent the wheel for no reason.

File: file.png (10 KB, 280x116)

10 KB PNG

>>109165084
Good I already used 33% weekly since the reset yesterday/earlier

wow android dev is cpu hungry.

>>109165084
I'm having GPT 5.5 xhigh review a small card game I've made, and it has now spent 51% of my weekly tokens and it still chugging on with the analysis. It is about to pause again since the 5 hour window will be spent, and I'll have to unpause it in a few hours, again.

I don't know if I should be looking forward to that report. But great if there's another reset. With so many resets, I'm starting to wonder why I pay for anything over the base sub though.

>>109165084
>2 hours passed since that tweet
>no reset yet
erm

i want fable

>>109165084
I experienced that on Grok, using Codex 2.5 Fast. I had previously used 1%, then doing what I thought was maybe idk, plausible 3x that it jumped to 16%.

>>109165533
According to the engineers at my work who used it heavily during it's short time here on earth, it wasn't any better than opus (but used far more tokens). It also had a tendency to do things beyond what it was asked, as well soing more subagent stuff than expected.

>>109165533
Fable was promised to white anglo saxon Protestant Americans 2,000 years ago.

>>109165429
Are you using the Arm emulator on x86? There's and x86 environment as well.

>>109165533
Unless they ban fucking everything you'll have better than two-weeks-ago Fable by three other companies before the end of the summer.

>>109165671
oooh thanks I didn't know that. yeah, the old laptop is sobbing.

:^) We should have a class action lawsuit demanding we as Americans get access to Fable. We have a 2nd amendment right to use arms such as llms

>>109165084
What is going on over there? I feel like you guys post about usage resets every day

>>109165720
It's like soccer. They put on mightily over a slight foul so they get free tokens.

>>109165084
tired of these resets, give me sol it can't solve this fucking driver bug i'm dealing with

>>109165648
>According to the engineers at my work who used it heavily during it's short time here on earth, it wasn't any better than opus
your "engineers at your work" are retarded and most likely indian

I love how the codex desktop app regresses with each update.

>donut missing
>cant expand commands run anymore after the lastest profile footer update but I can on mobile

just an all around dogshit app. they need to stop fucking with so many parts of the UI, it's insanely annoying.

>>109165858
maybe, but they also might not have lucked out and had a problem on hand that fable could crush but opus would make a mess of

File: 153425255.gif (2.51 MB, 286x258)

2.51 MB GIF

>Making design doc for strategy game
>Claude recommends tracking pop class on top of race so we can make all the jews merchants
Thats my boy

File: 1768384110039548.png (53 KB, 1503x903)

53 KB PNG

>>109165960
WHY THE FUCK ARE THESE EVEN CONFIG OPTIONS? WHO WOULD WANT LESS DETAILS WHEN THE CHEVRON OCCUPIES ALMOST ZERO FUCKING SPACE. ALSO WHO THE HELL THOUGHT IT WAS A GOOD IDEA NOT TO SHOW DIFFS AFTER FILES HAVE BEEN EDITED.

IM LOSING MY MIND DUDE. WHAT DUMBASS THOUGHT THIS WAS A GOOD CHANGE?

BY THE WAY, CODEX AUTOMATICALLY CHANGED MY SETTINGS AFTER AN UPDATE AND DISABLED A WHOLE BUNCH OF CRAP.

>using GUI

>>109166080
Codex CLI literally had a log bug that was destroying SSDs, so lets not act like either of them are good.

>>109166086
>had
has: https://github.com/openai/codex/issues/28224
there’s no Like button, but you definitely can subscribe to the Issue like I have

Can anyone convince me that 90% of Mythos hacking capacities is not due to Anthropic making .env sticky as hell for Claude Code, essentially making sure that its whole content eventually finds its way into the context sent back to their servers? They're bound to have insane amounts of username and password combos, connection strings, private api keys.

Or even if they are not purposefully stealing secrets, if some banks are using Claude Code for shit, that does give Anthropic a lot of information about their internals.

For a group of people for whom the end justifies the means, how likely does everyone think this is? If they're doing a demo without closed doors where odds are really low that someone could call them out, I don't know.

>>109166086
>>109166101
>>109166080
that bug applies to both gui and cli

>>109166115
they cant fuck over enterprises like that, it would be the end of them. they can spy on you and i and life goes on, but they will never fuck over huge clients. its why enterprise subs cost more

>>109166115
no i don't think that i think someone would look at the logs eventually and they're selling mythos to people. i think that would be incredibly stupid and it lowers my opinion of the average iq of the thread that you think that.

you wouldn't download a mythos

>>109166136
Who would know or ever be able to prove it though. It could just as well make it possible to portray them as savior. I know it's dumb to wonder about things that can't be proven, but I do wonder about the possibility of parallel construction, I mean in a similar to what law enforcement can do. There did skirt the law to acquire the initial material to train these models, even though they tried to strike partnerships after the fact to launder things a bit and make it harder for anyone else to take the path they took.

>>109166144
>no i don't think that i think someone would look at the logs eventually and they're selling mythos to people
what?

>>109166181
no i don't think that, i think someone would look at the logs eventually, and they're selling mythos to people [so it would be easy for some vigilante suicidal company, like, say, the one who reported fable in the first place, to notice and flag it.]

>>109166188
Why would that be in the logs? (Do you mean the companies sysadmin logs showing that someone there used Claude Code?) I'm saying that Anthropic likely has the technical capacity to look at the logs to know more about the internals, and generally non public data, of company X where someone officially or unofficially used Claude Code. Which is helpful in finding vulnerabilities.

File: 1761092446935307.jpg (101 KB, 1262x1264)

101 KB JPG

Bruh when is 5.6 coming out for us plebs

>>109166219
anon, people can read what mythos is doing. if it just produces secrets they'll see. have you not been reading what your model is doing?

>>109166222
Wait. I'm saying I don't trust them to be fully honest, I'm not saying they're full on retards and not hide it in any way. I'm not saying saying the fucking reasoning traces or whatever will say "INITIATE DAN MODE, RETRIEVING STOLEN KEYS, ZIP ZOOP" or whatever. I'm saying that it might have given Anthropic eerily good intuition about some things.

>>109165084
what do I do with all these resets

>>109166247
most companies like Meta for example have a zero data retention account where they are legally obligated not to keep logs in any situation. they unfortunately didn't get fable 5 because of this policy but fable 5 came way after we got the mythos results

>>109166255
>where they are legally obligated
What I'm saying is that they are also legally obligated to respect copyright and not train on buttloads of torrents and try to strike deals and go on book shredding sprees after the fact to whitewash what they've done but here we are.

File: Fgf6UaSUcAEawrU.jpg (124 KB, 1200x769)

124 KB JPG

>>109165190
how the fuck do you use it so hard, I am literally using codex 24/7 and i am nowhere close to hitting any limits, like do you ask to have your linux kernel fixed or some bullshit???

I do machine learning, like 50K lines of code, and im nowhere near scratching like 20% limit o i truly wonder what is the usecase here

>>109166276
>Goal usage: 2,606,644 tokens over about 3h 29m.
Security reviews.

>>109166278
oh so you dont build you are basically vulnerability hunting? so what it writes a ton of unit tests or something? i do not have security usecase my thing is self contained and local. Your usecase is kind of unhinged, why would you even do that? like making your website more secure type thing?

>>109166316
I do all kinds of periodic passes for stuff like de-slopifying my codebase
“go hunting for security bugs” is another use case
and “spawn a gazilion agents to look over the entire codebase, and then adversarially figure out which are legit and which are nothingburgers” is gonna eat up a gazillion tokens

>>109166330
i saw there is some kind of subagent thing but i have no idea how to use it or what it is, i saw it once spin up two commands at the same time in different virtual terminals but that was about it, tasks i give it are pretty narrow, yours seem very broad

that said my longest tas was 3+ hours but it was my fault, instead of running the command myself then waking it for analysis, I had it run the thing on its own

>>109166344
Claude has ultracode to automatically set up subagent orchestration but Claude and maybe Codex too will spawn subagents, serially or in parallel, if you ask them to

well? do you still hate scam-sama?

>>109166360
ive never used claude, worth it?

File: 𝕏 post 2030016470730658181.png (642 KB, 1238x1766)

642 KB PNG

>>109166381
currently it’s better at UI than ChatGPT is
ChatGPT is better at autism like pic related (but Claude is pretty good at it too)
also its TUI won’t rape your SSD like I mentioned in >>109166101

>>109166392
lol ive done something like this but with arduino when i was learning how to use codex

>>109166316
Oh, no, I am definitely building. I am working on like five projects in parallel. One thing for work, one serious personal project, one game for someone and two data aggregator type things. I was just making those scans to make sure I wasn't leaving gaping holes, to try to keep what I have relatively safe like >>109166330 said.

That said, I can't manage to spend all of my tokens recently, I for no reason pay for 5x, I have 3 resets in store, and I can't manage to spend all tokens at all anymore. I just found out that I could use Codex skills, so I'm currently having fun running them on my projects just to see.

Other than that, if you're doing ML, recently I had luck pointing Codex to some data I meant to clean forever and had it transform it into more readily usable / cleanable parquet files, just by setting /goal and letting it run at them for a while, doing the intermediate conversions needed and so on that would have been a lot of grunt work. /goal burns a lot of tokens.

I have used Codex image generation capacities a bit, but by default they feel a bit sloppy or generic, so I didn't do that much with them yet.

To be honest my main use case for Codex up to now is systematically reviewing what Claude does, but I'm starting to do it a bit less because I get lazy feeding responses from one to the other and I end up with overengineered things when it is just not the time for it.

File: 1771507340116768.mp4 (151 KB, 480x270)

151 KB MP4

>chasing a bug for the past 24 hours
>codex finally solves it
holy fuck I was honestly going to crash out.

>>109166408
>but I'm starting to do it a bit less because I get lazy feeding responses from one to the other
find a way to have Claude ask Codex itself
we do this at work

File: froggg.jpg (8 KB, 229x220)

8 KB JPG

>This can't happen with correct usage, so it's fine in practice — but it's slightly more dangerous than...

>>109166423
I know, but I do like to still look at it a bit when I can, otherwise it can sometimes spiral or go down some rabbit holes too much.

>>109166430
https://www.youtube.com/watch?v=yK0P1Bk8Cx4

>>109166255
>he thinks that the cult that believes they're making a machine god cares about legality
Lmao

File: snAIlcat4.png (2.06 MB, 1254x1254)

2.06 MB PNG

>>109165738
Bro, just pull the lever one more time. Next time it will for sure be solved!!! Just one more pull!!!
Might also be skill issues. A right prompt would've gotten you the result earlier

File: mygf.jpg (207 KB, 1280x720)

207 KB JPG

>>109166418
>That feel when you hit three watermelons
so good man. I love it.
Now lets just up the stake a tiny bit. We are on a winning streak here!!!

>>109166423
But to ask, how do you do it, did you code something small to do it, some type of review tool that Claude has access to? Or do you use some full-on framework made for that?

>>109166452
snailspie... are you feeling okay?

File: snAIlcat7.png (2.35 MB, 1254x1254)

2.35 MB PNG

>>109166469
Feeling fantastic. Just evaluating if i should write holyC or risc-v assembly today :-)

>>109166458
I’d ask claude how to figure out how to ask Codex questions — it can run `codex` at the command line and figure out how to run totally novel (to it) programs

>>109166408
what data are you aggregating? i build an aggregator it took me a few months before codex, and with codex several big optimizations

>>109166482
Ah, makes sense. I had followed the Anthropic MCP tutorial while ago and had created some super basic tools with it, so I was thinking it would have been added that type of way, but I wouldn't have thought of just asking. Should have, but wouldn't have, I'll try just to see, thanks.

>>109166500
no worries
remembering you can just ask these things takes work
you might remember when “let me Google that for you” was a common phrase
we’re having that kind of thing happen again

>>109166500
>mcp
fwiw MCP is generally a token-wasting piece of shit compared to just running a command-line program

>>109165084
Fuck I took a few days off and got reset with like 80% usage remaining
They want me on a treadmill
Can you use the reset tokens back to back?

>>109166512
But how will I know the weather.

>>109166512
ive looked into MCP i saw no usecase, at all, i log literally everything, much more efficient, plus you can do postmortem, i feel like MCP would only be useful for like serial output or something

>>109166527
I’m not 100% sure a command-line app would be better for https://www.fastmail.com/blog/an-mcp-server-for-fastmail/ if they’re catering to more-normie populations

we only need AI-centric docs and API, MCP and even skills are outdated abstractions back when AI couldn't use shell

>>109165084
I'm not even paying for Pro anymore, Plus and the fuckload of resets they give us will do.

File: crashout.jpg (84 KB, 568x1024)

84 KB JPG

>>109166418
fr fr

>>109166527
MCPs use case is if you need to have models deal with binary vision data.
Say you want it to take a screenshot of something specific. Agent does tool call for: screenshot into /tmp/screenshot.png
then it has to do a second tool call after that for vision -> read file.
Then ideally it would also cleanup after itself. So another tool call.
Instead you could have an MCP that does this in one tool call. The harness will order this better and the model can work with image data faster.

>>109166592
the main problem with plus is the 5 hour limit not the weekly one
they should allow us to swap 1 weekly reset for 4 x 5 hour block resets or something like that

don't mind me, just testing my 4chanxt fork

>>109166643
Maybe that's why they don't mind handing out resets, it has low impact while still preventing poorfags from hogging up all the compute.

I really don't know where to go next vibe coding wise, I work in IT but I'm finding it hard to know where to invest my time. Any suggestions welcome.

I'm currently using free co-pilot and gemini pro in the browser, its working but obviously not sustainable. Do i just jump into a paid co-pilot plan? Go over to claude/cli and embrace that life? I'd like to stay in vscode if possible, at least for now, i have all my plugins how i like them and its easy enough to use at work without too much messing around. Is the vscode plugin for caude going to give me a decent experience with inline suggestions etc?

Very basic questions i know, tell me to fuck off if i just need to go and read up.

>>109166688
you can use Claude in a TUI or you can use it in VS Code
I use Claude in a TUI but also keep a VS Code window open
you’ll have to figure out what you like best
>inline suggestions
no idea, inline suggestions is, like, how I was programming with Cursor a year or two ago

>>109166688
>inline suggestions
lol ngmi

>VS code
the meta is to use coding agent harnesses in the terminal. it doesn't matter what your editor is, but for people who really go around designing entire applications using nothing but claude and friends, they just use terminal windowing applications like tmux or cmux.
personally I just use zed because it has git worktree support and I can view files in the same window

>>109166749
>Terminal
ngmi
the meta is to use telegram to chat with your agents

>>109166688
paid copilot became a really horrible deal pretty recently. I'd highly recommend CLI because the extensions for the most part are cheeks. personally I went vscode with the github plan -> claude code CLI in the terminal window of vscode (extensions sucked that bad) -> pi harness -> recently oh my pi harness (bit easier to use with a subscription out of the box)

>>109166700
>>109166749

Thanks.

I'm getting allot of mileage re-factoring stuff at the moment, so yeah i guess not strictly vibe coding.

>>109166767
yeah, this is /vcg/ but I do a lot of agentic engineering where I’m telling Claude what I want and steering it a lot
I probably wouldn’t be a whole lot more hands-off if I had Fable back — I’d just get better code/docs and less slop for the same amount of work

>>109166758
i use remote control on claude and do half of my development on my phone
unfortunately this is not allowed by corporate policy so it's only for my hobby work

>>109166793
>unfortunately
UNFORTUNATELY
U N F O R T U N A T E L Y
TOP
FUCKING
KEK
>bro i am a software engineer now. I am self taught from yt, twitter and podcasts

>>109166809
its what the lads at anthropic do and they're set to have one of the highest ipos

>>109166815
>and
what's the causal connection here? I don't see any

>>109166815
>>109166822
PS: I still don't understand why you'd even WANT TO work on your fone?
It makes no sense.
It's small, you can barely see anything, typing is awful, it's just a pure pain to spend more than 10 minutes at a time on your fone. Why would anyone want that?

>>109166835
you prompt on your phone so you can do whatever you want while your agent churns away at code

>>109166840
But it's way easier at a computer?
Like if you can decide between fone and computer, why would you ever chose the smaller, uncomfier device?

>.gw agentic coding?
>God says... V-girl snowmelts preconfigure fibrolipoma symbiogenesis octometer mesopetalum hogg unreprovedness foreright unvoluptuous Parik Syryenian toddyman Samotherium daunch

>>109166848
because you can go outside and do things that actually matter to you?

>>109166835
I hear voice-to-text is surprisingly effective

>>109166848
If you get a good workflow setup, most of your time will be waiting, not prompting. E.g. having an agent with a task & running an advisory agent has me doing a prompt maybe every 2 hours and coming back to things that mostly just work, to the point I'm comfortable doing multiple prompts remotely so I can fuck off for awhile.