SSE @codecentric. OSS. Enterprise Java. Maintainer of https://teapaw.org/t.co/UGCGylSwkT. Member of JUG Hamburg and iJUG. PC @JavaLandConf. Tweets are my own.
Lüneburg, Germany
Joined July 2014
- Tweets14K
- Following253
- Followers986
- Likes8.8K
Wenn du selbst etwa 15.000€ netto monatlich bekommst, deine Freunde und Familie in NGOs auf Steuerzahlerkosten unterbringst und dann forderst, dass andere die weniger verdienen mehr Steuern zahlen sollen. Dann bist du in der SPD angekommen.
Bennet Schulz retweeted
"I'm willing to go bankrupt rather than lose this race." Google co-founder Larry Page on the AI race.
Bennet Schulz retweeted
Man kann dem Bundeskanzler kein Wort glauben. Sein letztes noch bestehendes Versprechen hat er heute gebrochen.
Bennet Schulz retweeted
Just released: lightmetal
GPU LLM inference on Apple Silicon: a single Java 25 executable JAR, zero dependencies.
Binds Metal-enabled libllama.dylib through the Foreign Function & Memory API. Runs Mistral and Gemma GGUF models locally.
- Anthropic-compatible /v1/messages
- OpenAI-compatible /v1/chat/completions
- Embeddable via ServiceLoader<BinaryOperator<String>>
- No Python, no Ollama, no container
👉 github.com/AdamBien/lightmet…
#Java #LLM #AppleSilicon #bce #airhacks
Bennet Schulz retweeted
I'm hiring product and AI engineers to work on the GitHub Copilot App.
If you're the right person to invent future of work, reach out to me with some proof.
If you work at GitHub/Microsoft and this sounds appealing to you, reach out to me internally with similar artifacts.
Bennet Schulz retweeted
“Software engineers who don’t know how to use AI coding agents will fall behind.”
No.
AI coding agents are not the hard part of building software. They’re a simple tool. You can learn to use them in a few days.
The hard part is knowing what to ask for.
Knowing whether the answer is any good.
Knowing when the code is brittle, overcomplicated, insecure, or just plain wrong.
It’s the judgement required to use them well.
That’s software engineering. And that takes decades to learn properly.
The people at risk aren’t engineers who haven’t mastered using coding agents yet.
It’s people who only know how to prompt one.
Bennet Schulz retweeted
🦔Hackers stole high-profile Instagram accounts including the Barack Obama White House account, the Chief Master Sergeant of Space Force, and Sephora by typing a sentence into Meta's AI support chatbot asking it to change the email on the target account. The bot sent a verification code to the attacker's email, the attacker entered it, and got a password reset. The exploit worked for months.
In March, Meta rolled this AI support out to all Facebook and Instagram accounts with password reset authority and no option to escalate to a human. Their own blog post promised the AI could "prevent an account takeover."
My Take
Meta just cut 2,212 people from Menlo Park, gutted its core engineering teams, and handed account security to a chatbot that couldn't tell the difference between an account owner and a stranger. The hackers didn't need a sophisticated exploit. They typed a request and the bot complied.
An infosec professional in the 404 Media thread made a point worth repeating. LLMs are not deterministic. You can't secure them the way you secure traditional software. You add a second AI layer to check the first one and you've just moved the vulnerability up the chain while adding more attack surface. There's no configuration that turns a language model into a security system.
Meta gave the bot password reset authority because it was cheaper than paying humans to do support. The Obama White House Instagram got taken over because a chatbot said yes to a stranger. That's what it costs to replace your support team with software that agrees with everyone.
Hedgie🤗
404media.co/hackers-simply-a…
Bennet Schulz retweeted
New GitHub Copilot pricing is here. And it requires new usage behaviour.
Left (light):
- Ask mode with Auto model (10% discount)
- 0.8 AI credits
Right (dark):
- Agent mode with Sonnet 4.6
- 4.3 AI credits
Same prompt and same useful answer. 81% cheaper.
Bennet Schulz retweeted
As you write code by hand, you intrinsically review it. As you gain experience by writing (and thus reviewing), your code gets better. Your reviewing skill gets better. You spot your own mistakes faster, and address them as you write and rewrite the code.
AI written code is a code dump. Reviewing AI code dumps is as hard as reviewing a very long PR someone sent your way.
Expecting humane understanding of AI written code is unfathomable in many cases. Only using AI to review AI generated code seems plausible.
Creator of C++, Bjarne Stroustrup:
AI-generated code isn't ready — it generates more bugs, more bloat, more security holes, and is nearly impossible to validate
"senior developers are already retiring rather than deal with it"
The problem is that even a small prompt change can shift the entire codebase in unpredictable ways
Bennet Schulz retweeted
Free tickets 🎟️ + real #Java conversations!
Join us at @JCON_Conference Europe 2026, Apr 20–23 in Cologne, Germany
Java • Modernization • Performance • AI
#JCON #AI
devblogs.microsoft.com/java/…️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️
Bennet Schulz retweeted
EBITDA (2026 edition): Earnings Before Iran, Tariffs, and Donald Announcements.
Bennet Schulz retweeted
Davos is basically all the people who have created the problems, denied their existence and then called us names for pointing them out... explaining what the problems are and how we desperately need to fix them.
German Chancellor Merz admits:
Both Germany and Europe have wasted incredible potential for growth in recent years by dragging their feet on reforms and by unnecessarily and excessively curtailing entrepreneurial freedom and personal responsibility.
We must substantially reduce bureaucracy in Europe. The single market was once created to form the most competitive economic area in the world.
Instead, we have become the world champion of overregulation. That has to end.