yonder

8.3K posts

yonder banner
yonder

yonder

@yondercode

milady 🦭

Bergabung Ekim 2010
1.8K Mengikuti3.1K Pengikut
Sudo su
Sudo su@sudoingX·
a corporate salesman on an openai paycheck tells you local models aren't there yet. an influencer selling you an API wrapper calls the local AI community on X "cancer." meanwhile we're out here modding communities, helping strangers debug their configs at midnight, fighting spam, pushing open source, and doing it all for free. these people don't want you running models on your own hardware. they want you as a customer. every local install is revenue they lose. every migration from their bloat is a subscription cancelled. don't let corporate noise and engagement bait merchants convert you into their recurring revenue. buy a GPU. compile from source. own your thinking. the community they call cancer is the same community that will help you get started for free while they charge you per token.
English
70
83
1.1K
110.2K
yonder
yonder@yondercode·
@Teknium intel qwen 3.5 122B int4 autoround, run with the eugr vllm container he has a recipe
English
0
0
1
94
Teknium (e/λ)
Teknium (e/λ)@Teknium·
Just got an Nvidia Spark setup. Hermes Agent installed without any issues. Now lets see what model it should be powered by 😉
English
37
7
294
14.4K
Dennison
Dennison@DennisonBertram·
@arafatkatze I think their goal is different no? They want to have local models that run on local machines.
English
1
0
14
3.2K
yonder
yonder@yondercode·
@jamonholmgren multimodal LLMs project image patches into the same vector space as text, treating them as a stream of 'image tokens' that the model processes just like other text tokens, pretty cool really
English
0
0
3
1.5K
yonder
yonder@yondercode·
@ziwenxu_ @DatSciX asus ascent gx10 is $3k, it has identical chip, better cooling, but 1TB NVMe instead of 4TB
English
2
0
1
94
Ziwen
Ziwen@ziwenxu_·
@DatSciX Who said not haha.. But it's 5k lmao
English
2
0
1
204
yonder
yonder@yondercode·
@NeonThunderX @stacy_muur vibecode, follow and experiment with the <new thing>, accumulate btc best time to be a neet chud
English
0
0
2
17
Stacy Muur
Stacy Muur@stacy_muur·
Question to all OGs who survived the 2018 & 2022 bear markets What's the best advice you have for the newbies going through their first bear market now?
English
417
23
491
148.9K
yen
yen@yen8902·
🫍
ART
13
5
97
41.1K
yonder
yonder@yondercode·
@PaulSolt im usually running on high, since i think intelligence wise: 5.2 medium > opus > 5.2 high im running it on vscode codex extension though, maybe that matters?
English
0
0
1
48
Paul Solt
Paul Solt@PaulSolt·
@yondercode I don't see the slowness. I think you might have a bad agent file, or maybe you're running on X high. It's really fast on medium and high.
English
1
0
0
59
yonder
yonder@yondercode·
@daniel_nguyen 5.2 is also extremely lazy but that gives it the aura > asks it to optimize some parts of the code slop written by opus > takes almost an hour to think (xhigh) > doesn't even bother running tests / bench, expects the human to do it > all tests pass and massive perf gain
yonder tweet media
English
1
1
2
314
Daniel Nguyen
Daniel Nguyen@daniel_nguyenx·
Codex 5.2 works so well for me than I haven’t really spent much time with Opus 4.5. I did a couple of tasks with Opus and the results were decent. But I still need to actively steer the conversation a lot. With Codex, I usually just chat until I’m happy with the implementation plan. Then I leave it to do the work, start other tasks, come back to review and test. It’s slow, but most of the time it works correctly without any further edit. For challenging tasks like crash log analysis, GPT 5.2 Pro is very good at finding the root cause. But yes, It’s very very slow. Are you using both Opus and Codex, which one works better for you?
English
72
14
401
29.8K
yonder
yonder@yondercode·
@allgarbled best time to be a neet unironically
English
0
0
8
8.3K
gabe
gabe@allgarbled·
I think probably the best job in the world you could have right now is unemployed software engineer
English
202
209
7.7K
700.3K
♡ Charlotte Fang 🪲 Crown Prince ❀ LOVE HEALS 💞
milady
vitalik.eth@VitalikButerin

I agree with maybe 60% of this, but one bit that is particularly important to highlight is the explicit separation between what the poster calls "the open web" (really, the corposlop web), and "the sovereign web". x.com/tom777kruise/s… This is a distinction I did not realize until recently, and I must admit the bitcoin maximalists were far ahead: a big part of their resistance to ICOs, tokens other than bitcoin, arbitrary financial applications, etc was precisely about keeping bitcoin "sovereign" and not "corposlop". The big error that many of them made was trying to achieve this goal with either government crackdowns or user disempowerment (keeping bitcoin script limited, and rejecting many categories of applications entirely), but their fear was real. So what is corposlop? In essence, it is the combination of three things: * Corporate optimization power * An aura of respectableness of being a company with sleek polished branding * Behavior that the exact opposite of respectable, because that's what's needed to maximize profit Corposlop includes things like: * Social media that maximizes dopamine, outrage, other methods of short-term engagement, at the expense of long-term value and fulfillment * Needless mass data collection from users, often followed by managing it carelessly or even casually selling it to third parties * Walled gardens charging monopolistic high fees and actively preventing people from even linking to other platforms * Hollywood releasing the 7th sequel to some tired franchise, because that's the most risk-averse thing to do * Every corporation that rallied around slogans of diversity and equity and the need to overturn society to fight racism in 2020, and then publicly mocked those causes for engagement in 2025 This is all digital corposlop; there are big and important analogues to this in the physical world too. Corposlop is soulless: trend-following homogeneity that is both evil and lame #how-we-fear-big-business" target="_blank" rel="nofollow noopener">vitalik.eth.limo/general/2025/1… These are things that appear to serve the user, but actually disempower the user. I have many qualms with Apple, but aside from their monopolistic practices, they actually have many non-corposlop traits. They serve users not by constantly asking "what do users want this quarter", but by having an opinionated long-term vision. They have a strong emphasis on privacy. They resist and create trends rather than following them. I just wish they could take the brave step of ending their monopolistic practices and switch to an open source first strategy. It may damage their market cap, but man must live for something higher than market caps. Zac from Aztec was also early to recognize the importance of this, with a post that is on the whole very pro-freedom, but at the same time does not shrink back from labeling what is essentially corposlop a primary enemy, even when it does not violate the libertarian non-aggression principle. x.com/Zac_Aztec/stat… In 2000, the understanding of "sovereignty" largely focused on avoiding the iron fist of government. Today, "sovereignty" also means securing your digital privacy through cryptography, and securing your own mind from corporate mind warfare trying to extract your attention and your dollars. It also means doing things because you believe in them, and declaring independence from the homogenizing and soul-sucking concept of "the meta". These are the kinds of tools that we should build more of. Build tools like: * Privacy-preserving local-first applications that minimize dependence on and data leaks to third parties * Social media platforms and tools that let the user take control of what content they see. Appeal to people's long-term goals, not short-term impulses * Financial tools that help users grow their wealth, and do not encourage 50x leverage or sports betting or taking out a loan to pay for a burrito * AI tools that are maximally open and privacy and local-friendly, and that maximize productivity from merging the power of human and bot, rather than encouraging the user to sit back and let the bot do all the work, so they learn nothing * Applications, companies, and physical environments that take an opinionated view on the kind of world they want to see, and have an opinionated culture * DAOs that can support organizations and communities that steadfastly pursue a unique objective, and do not all get captured by the same groups. Privacy-preserving and non-tokenholder-driven voting can help here Be sovereign. Reject corposlop. Believe in somETHing.

English
87
62
544
64.4K
yonder
yonder@yondercode·
@webdevcody yeah i am starting to use 5.2 for the solution and paste it to opus for the grunt work, problem for 5.2 is it is really damn slow
English
1
0
2
440
Charlie Lamb
Charlie Lamb@charlietlamb·
google dropped antigravity and still don't know one person that actually uses it
English
287
2
315
55.3K
Alula
Alula@__alula·
@errors_here kinda agree but it's only fun until you have to debug
English
25
3
367
74.2K
yonder
yonder@yondercode·
@HaanzeR @antigravity but problem with antigravity though the agent often crashes, it's just a click to tell it to continue but it's such a hassle and the UI is pretty buggy and laggy
English
1
0
1
69
Google Antigravity
Google Antigravity@antigravity·
As we balance giving the best possible quotas and maintaining fairness between users, especially under incredible demand, we will be establishing generous weekly limits for all models. This will only affect a minority of Google AI Pro users. These limits do not apply to Google AI Ultra, which continues to be the best plan for power developers!
English
691
187
3.6K
944.6K