Sabitlenmiş Tweet
The new open source LLM from @OpenAI feels unreal, insanely fast for a locally running 20B model 🚀 and quality looks good, too. First response: Horizon Beta @ OpenRouter (#GPT5?), second: local #gptoss20b 🥰
English
Jan Ole Suhr
63K posts
Jan Ole Suhr
@janole
Hands-on, open-minded. Software Dev & Product Engineer (since 1986), Entrepreneur by Heart, CPTO. Creator of #Gravity—the Social Net App for Nokia (1M users)
Berlin, Germany Katılım Temmuz 2007
129 Takip Edilen7K Takipçiler
Jan Ole Suhr retweetledi
@geerlingguy We've disabled it already. Basically it was giving product tips which was kinda ok on Copilot originated PR's but then when we added the ability to have Copilot work on _any_ PR by mentioning it the behaviour became icky. Disabled product tips entirely thanks to the feedback.
English
That reminds me of a related trait: being able to code completely offline. No Google, no Stack Overflow, just man pages ... 😅 #oldschool
Gergely Orosz@GergelyOrosz
Devs who can code also WITHOUT AI as well looking to became 10x more valuable They are the ones who won’t panic or be idle when their Claude quota runs out… So much for all the advice on how learning to code is not worth it any more…
English
Jan Ole Suhr retweetledi
Remembering Leonard on what would have been his 95th Birthday 🎂
English
Claude acting up again today. In the future, we'd hopefully get some sort of post-mortem like we get from AWS, Cloudflare, ...
English
Jan Ole Suhr retweetledi
LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below
English
Something is up with Claude Code usage today. $200 Claude Max, 0%, 52% to 62%, then 68%, 76% and 84% in 5-hour rolling window in the time it took me to write this tweet. WTF, @AnthropicAI? I'm working on one GitHub PR for regression testing. Not folding proteins to cure cancer.
English
@BradGroux @AnthropicAI Same over here. To make matters worse, the generated code was overwhelmingly bad today 😵💫
English
The more I think about it, "disposable code" feels just like a variant of (Throwaway-)Prototyping which we got taught in university in the early 1990s 😅
English
@emmaedeh @addyosmani Missed that. Great post. Thanks for mentioning it 🙏👍😎
English
Jan Ole Suhr retweetledi
@janole This is the biggest dilemma right now!
@addyosmani captured it brilliantly.
addyosmani.com/blog/comprehen…
English
I think this is a real problem: we are now generating code in a week that used to take months or more before. There is no way you can understand these codebases that are growing in an explosive manner like you used to understand codebases that grew „organically“.
It is like being pushed into a new mega-project every other day or so.
We need some new ways or schemes to handle this kind of software development …
Some ideas like creating good specifications up front sound scarily similar to the waterfall model … (That’s not necessarily bad, waterfall is still used in a lot of places although usually branded as Scrum or „Agile“ 😱😂)
I like the way we can now refactor or even rewrite large portions of a codebase with little effort („Disposable Code“), but I am not sure how helpful this is: in the end, I assume, we are looking for a stable (perfect?) codebase. But maybe that’s just the old way of thinking about software engineering.
If you realize it’s less than a year that we can even generate „proper code“ with LLMs, we are just at the beginning of a fundamental shift in software engineering.
camsoft2000@camsoft2000
I’m getting to the point with one of the projects I work on where the complexity of AI slop is becoming a real issue. While I can still happily prompt the agent to add x feature and it will do so and it will likely work perfectly, the code is just getting too complex and fragmented. Agents love to copy and paste and keeping patterns DRY is a real challenge. The agent will start diverging all those copy and pastes until you’ve got loads of similar but slightly different blocks of logic. Again it all still works and solves the problem I’m after. But I just can’t get any kind of consistency anymore, the code is a mess and I just don’t have a handle on it. I want a clean unified architecture but agents just code with tunnel vision. The project is now too big and complex for an agent to fully reason with and too big and complex for me to reason with. The only real solution is a complete rewrite. Maybe this is the way things will go. Code will just become disposable. I don’t really want to care about the code and to be honest I don’t but I do care about consistency and maintainability and the AI slop is hurting those very things I do care about. I know some will say “I’m holding it wrong”, use x,y,z skill, tool whatever and already use tools and anti slop skills, plans, docs, etc but the outcome is the same. Vibe coding something into existence is truly magical. But turning it into a mature product with months of iterations is painful. I can’t even hand code this thing because I don’t understand the code anymore and I’m too lazy to try and code myself because I’m addicted to AI. So what’s the solution, either start again and accept that’s just the way we have to roll, or just carry on fighting the slop and accept each new feature will take longer to implement than the last. I’m tired. I’m addicted.
English
@badlogicgames I have the weird feeling it’s all because the LLMs learned from millions of „1. this is wrong … 2. this is how you do it right“ samples and now it needs at least one wrong step before it can reveal the solution 🥶🧙♂️😂 #magic
English
@shadcn Self-made app, start and use on web (regardless of desktop or mobile), connected to a hosted runner and a local runner using Vercel AI SDK + codex adapter for programming. Usually back and forth: start on mobile, continue on desktop, check progress on mobile, …
English
Tried to find out details about some upcoming remote access feature in #Codex. Instead the search came back with an „automated approval system via subagents.“ That’s brilliant! Have been using a very similar thing in a private project and it just works great.
English
It’s so hilarious to see an LLM steering/overseeing/interacting with a sub-agent. It’s often like: „The agent wants to write to a file. I approve …“ - „The agent is finished. Great work by the agent. Here’s the summary…“ 😂😂😂
English
Jan Ole Suhr retweetledi
