Rasmus Schultz

13.8K posts

Rasmus Schultz banner
Rasmus Schultz

Rasmus Schultz

@mindplaydk

web developer since 1998 I hate that I'm on this platform, but apparently developers are stuck in purgatory for eternity 👿

Odense, Denmark Katılım Aralık 2009
190 Takip Edilen702 Takipçiler
Sabitlenmiş Tweet
Rasmus Schultz
Rasmus Schultz@mindplaydk·
half my feed is now random memes I have no idea what they're doing with the feed or why, but I'm out this app is a complete waste of time I'll check back in a month or so bye
Rasmus Schultz@mindplaydk

I am almost over this wtf did they do? my feed is completely random junk practically nothing from the developers I've been following for more than a decade if they don't fix this in the next couple of days, I am OUT might as well be on BSky, their shitty feed is BETTER now

English
0
0
0
128
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@mattpocockuk @JayD0ubleu there's a good chance it's just random though? on some tasks, I've found that literally just stop, clear and retry, with the same exact prompt, can often fix it I've repeated this experiment enough times to not even bother editing the prompt on the first retry 😊
English
0
0
0
11
Matt Pocock
Matt Pocock@mattpocockuk·
@JayD0ubleu Maybe - and this is always a possibility - but I do doubt it on this occasion. Especially since I cleared the context and it solved the bug in 80K tokens.
English
2
0
5
2K
Matt Pocock
Matt Pocock@mattpocockuk·
Doing some experiments today with Opus 4.6's 1M context window. Trying to push coding sessions deep into what I would consider the 'dumb zone' of SOTA models: >100K tokens. The drop-off in quality is really noticeable. Dumber decisions, worse code, worse instruction-following. Don't treat 1M context window any differently. It's still 100K of smart, and 900K of dumb.
English
152
59
1.2K
154.5K
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@p3x1967 @rohanpaul_ai Neurophos is shipping working evaluation systems this year they can manufacture on existing fabs, so the possibility that they'll ship at scale in 2-3 years looks real hallucinations and Idiocracy, yeah, no one has any answer to that 😌
English
0
0
0
21
Craig
Craig@p3x1967·
"...in the next 2-3 years" Meaning in the next 10 to 20. Meanwhile, the hallucination problem with LLM's can never be fixed. It's engineered in at the most fundamental level. And none of this changes the fact that using AI is making humans dumber, less innately creative, less resilient...
English
1
0
0
7
Rohan Paul
Rohan Paul@rohanpaul_ai·
Brilliant economic paper, directly models the "Structural Jevons Paradox" happening right now in the AI industry. The cost of running an LLM is dropping, but total computing energy is exploding anyway. It mathematically proves that as the unit cost of digital intelligence and coding drops, the aggregate demand for complex AI agents and the infrastructure to support them surges exponentially, creating a massive new downstream ecosystem that requires human management. Reveals a massive paradox where dropping the price of AI usage does not save money, but instead encourages developers to build vastly more complex agents that eat up exponentially more computing power. Because of this relentless progress, small companies building simple applications on top of these models get completely crushed as the core AI naturally absorbs those exact same features over time. They also discovered a brutal dynamic where a perfectly working LLM becomes economically worthless the moment a competitor releases a smarter version. Ultimately, the researchers prove that this combination of massive computing costs and the need for constant user data naturally pushes the entire AI industry toward an unavoidable monopoly. --- arxiv. org/pdf/2601.12339v1 "The Economics of Digital Intelligence Capital"
Rohan Paul tweet media
Rohan Paul@rohanpaul_ai

Citadel Securities published this graph showing a strange phenomenon. Job postings for software engineers are actually seeing a spike. The graph here is short term but still it's super interesting and really strange. Is it Jevons paradox at play. When AI makes coding cheaper, companies actually may need a lot more software engineers, not fewer. When software is cheaper to build, companies naturally want to build a lot more of it. Businesses are now putting software into industries and tools where it was simply too expensive before. --- Chart from citadelsecurities .com/news-and-insights/2026-global-intelligence-crisis/

English
40
159
592
55.8K
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@p3x1967 @rohanpaul_ai there already are low cost options MiniMax M2.5 is very cost effective there will be even better and cheaper ones in the future plus the hardware will get cheaper and way more efficient - with photonics, maybe 100x cheaper in the next 2-3 years watch 😊
English
1
0
0
39
Craig
Craig@p3x1967·
@rohanpaul_ai Nobody is paying anywhere near what it costs to actually deliver this. There won’t be any such thing as a low cost LLM in the future.
English
1
0
2
59
Rasmus Schultz
Rasmus Schultz@mindplaydk·
"reasoning" has no right to be called "reasoning" I know it "works" for certain tasks, but it has nothing to do with reasoning - the model just generates "more words" in the hopes of fuzzily stumbling into some useful activations some percentage of the time
Rasmus Schultz tweet media
English
0
0
1
27
Rasmus Schultz
Rasmus Schultz@mindplaydk·
tbh Claude Code is hopelessly mismanaged and a complete mess #issuecomment-3909294004" target="_blank" rel="nofollow noopener">github.com/anthropics/cla… opened two detailed issues and the bot just closes everything without anyone ever even looking OpenCode is vibing pretty hard rn, but at least they haven't completely given up on governance 😌
English
0
0
1
82
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@CFDevelop it depends on the task, of course. but in some cases, yes - for example, you can't know the decision making behind certain details except through discovery. this was always a limitation of code reviews - it's why pair programming is better than code review for certain tasks.
English
0
0
0
17
Christian Findlay
Christian Findlay@CFDevelop·
@mindplaydk No. You’re saying the only way you can learn or trust code is by actually writing it?
English
1
0
0
10
Rasmus Schultz
Rasmus Schultz@mindplaydk·
you might see nothing new here, but the fact that he *stays in flow* 👀 I've never seen anybody work with AI in this way - most people wait for the AI to complete a step, then review. this feels more like coding, accelerated 🤔 youtu.be/ws9zR-UzwTE
YouTube video
YouTube
English
1
0
0
126
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@CFDevelop code reviews are mostly ceremony. you do not gain the same level if understanding, insights, trust, or ownership, from just reading code, as you would from writing it. surely you must know this, having worked on software teams for many years? 🤷‍♂️
English
1
0
0
12
Christian Findlay
Christian Findlay@CFDevelop·
@mindplaydk You’re theorising about stuff that doesn’t matter What matters is the quality of the product and tests at the end of the day. And yes, I’m including the code as part of the product The code is right there. You can review it at any time
English
1
0
0
6
Rasmus Schultz
Rasmus Schultz@mindplaydk·
Open models are over just as fast as they arrive. I've tried two of these (Minimax M2.5, Kimi K2.5) but I think I'll just skip GLM5 and withdraw from the pursuit of open models for another 3-6 months. There is real progress, but never what YouTubers promise every single day. 🙄
Rasmus Schultz tweet media
English
0
0
1
126
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@CFDevelop I'm not saying every problem benefits from careful exploration. just that some do - and it's often difficult to know in advance which problems will or won't benefit, because you don't know what you don't know. it's still a matter of weighing the value of understanding vs speed.
English
1
0
0
8
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@CFDevelop specs can't predict everything. many tasks require you to engage with the actual code - the insights won't surface until you do. it depends on the task, of course. but you can't always know the best approach before you're in the middle of it. AI can skip valuable process 😌
English
2
0
0
8
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@CFDevelop without understanding and good decision making? I rest my case. 😌
English
1
0
0
12
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@airesearch12 maybe when we get photonics I hear test devices will ship this year so probably available for consumers in 5 years or so until then, Big AI owns your ass * owns everyone's ass ** still owns everyone's ass in 5 years for anything large enough to be truly useful 😌
English
0
0
0
18
Florian S
Florian S@airesearch12·
* ok, not consumer grade laptops, super expensive laptops with very expensive NVIDIA GPUs, but still ** probably benchmaxxed, but still
English
6
2
60
5.6K
Florian S
Florian S@airesearch12·
There it is. We have 800K context Sonnet 4.5 at home, on consumer grade laptops.
Qwen@Alibaba_Qwen

The Qwen3.5 series maintains near-lossless accuracy under 4-bit weight and KV cache quantization. In terms of long-context efficiency: Qwen3.5-27B supports 800K+ context length Qwen3.5-35B-A3B exceeds 1M context on consumer-grade GPUs with 32GB VRAM Qwen3.5-122B-A10B supports 1M+ context length on server-grade GPUs with 80GB VRAM In addition, we have open-sourced the Qwen3.5-35B-A3B-Base model to better support research and innovation. We can't wait to see what the community builds next!

English
31
76
2.1K
246.1K
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@CFDevelop you're probably still just a code reviewer then sounds overwhelming the point is to stay engaged with what you're doing, gaining understanding and making valuable decisions
English
1
0
0
16
Rasmus Schultz
Rasmus Schultz@mindplaydk·
why is "100% Prettier compatibility" always the first thing advertised by every alternative to Prettier? Prettier is awful - cramming as much code into every line as possible is not a "gold standard" that formatters should chase 😮‍💨 please try dPrint! and spread the word 😊
English
0
0
0
65
Rasmus Schultz
Rasmus Schultz@mindplaydk·
you know how you click in VS Code or Brave on Linux, and for some reason it drags instead of clicking? no, it's not a hardware defect, it's this: bugs.launchpad.net/ubuntu/+source… first reported in 2010 how does something this small and basic just get ignored for 15 years? 😟
English
0
0
0
112
Rasmus Schultz
Rasmus Schultz@mindplaydk·
looks like JPEG-XL support is once again under way in Chromium 🤩 issues.chromium.org/issues/4629193… if you recall, it was removed over politics and disputes over patent grabs by Microsoft and Google 😒 Microsoft has already commented and asked that the feature be made optional 🤦‍♂️
English
0
0
0
52
Rasmus Schultz
Rasmus Schultz@mindplaydk·
@penpotapp yes, but a few things are still lacking mainly border-left/right/top/bottom, which isn't supported by PenPot yet that, and styles (shadows, typographies) aren't being imported properly both issues are tracked on the Github already 😊
English
1
0
1
18
Rasmus Schultz
Rasmus Schultz@mindplaydk·
calling it: @penpotapp will eat Figma 😄 github.com/penpot/penpot-… this isn't 100% yet, but it's already close and the design tokens system in PenPot is so much more powerful than the mess of variables and styles in Figma incredible progress! I am so excited for this 😄
English
1
0
2
145

Keşfet

@mattpocockuk @JayD0ubleu @p3x1967 @rohanpaul_ai @CFDevelop @elonmusk @BarackObama @taylorswift13