matthiasneumayer.com
Founder
Building stuff with @dimarubanov.com
🦊📚 oscarstories.com
🤖📖 fragdaspdf.de | https://www.intellipaper.ai
🔥5️⃣ branding5.com
LL.M. (law) & BA (film) | Former Film director | Trainee Lawyer | Software Developer
1,205 posts
4,244 followers
1,063 following
Regular Contributor
Active Commenter
comment in response to
post
Like magic ❤️
comment in response to
post
Thx mate!
comment in response to
post
true! but I think because MCP is basically "kind of" RAG, if training data contradicts the docs, there still could be problems
comment in response to
post
Congrats 👏
comment in response to
post
Yes especially when it comes to architecture, gemini and others tend to write a lot of code for simple functions. And optimising is not really in their ballpark
comment in response to
post
Absolutely true!
comment in response to
post
That helps, but can create friction, especially when the old and new information don’t quite align.
Now imagine a future where most code is written by AI.
And platforms like StackOverflow fade out.
Where does high-quality training data come from then?
comment in response to
post
No surprise that today’s AI models are really good at coding.
But here’s where things get tricky:
Libraries change. Frameworks evolve.
And the model doesn’t know about any of that.
Tools like Cursor use RAG to pull in the latest info from docs, but the model itself is still trained on older data.
comment in response to
post
Hahah I missed your posts ❤️
comment in response to
post
💯
Gibt auch zahlreiche papers dazu: arxiv.org/abs/2409.05746
comment in response to
post
Danke dir, das freut uns sehr 😊
comment in response to
post
We'll enter a world of super shitty software, followed by a wave of AI bug-fixing startups. Then companies will provide humans who fix AI bugs. Then AI-driven bug-fixing talent recruitment platforms. Eventually people will start hiring real developers who can actually code. Full circle.
comment in response to
post
Maybe @developer.microsoft.com ?
comment in response to
post
Not true, for AI search (chatgpt) it is now sadly important
comment in response to
post
Crazy! And the support is basically non existent.
comment in response to
post
@mscommunity.bsky.social @bingg.bsky.social
comment in response to
post
Funny thing is, we've had great experiences with Microsoft Azure. Which makes this contrast all the more frustrating.
👉 If anyone knows someone on the Bing team or has dealt with similar issues, I'd love to connect. Talking to an actual person would make a big difference here.
comment in response to
post
The site has a domain rating of 46, and Google indexes it just fine. But Bing keeps saying it’s “not allowed” to index it, with no clear explanation.
After waiting 3 months for a reply, the response we finally got didn’t help at all. No insight, no next steps, just a generic message.
comment in response to
post
Manchmal wär ich auch gerne so verblendet und ohne jegliche Bodenhaftung, muss irgendwie angenehm sein
comment in response to
post
Gute Besserung und erhol dich gut!
comment in response to
post
I mean yes for prototyping this is cool, but I agree production will kill many of thise
comment in response to
post
We built a tool that automatically replaces DEI terms with similar-looking Unicode characters, so terms like 'diversity' or 'LGBTQ' remain readable to humans but are not recognized by censorship systems. So agencies can circumvent Trumps agenda
www.intellipaper.ai/en/other-too...
comment in response to
post
We built a tool that automatically replaces DEI terms with similar-looking Unicode characters, so terms like 'diversity' or 'LGBTQ' remain readable to humans but are not recognized by censorship systems.
www.intellipaper.ai/en/other-too...
comment in response to
post
We built a tool that automatically replaces DEI terms with similar-looking Unicode characters, so terms like 'diversity' or 'LGBTQ' remain readable to humans but are not recognized by censorship systems.
www.intellipaper.ai/en/other-too...
comment in response to
post
Thx for sharing :)
comment in response to
post
We built a tool that obfuscates those phrases to avoid detection and censorship: www.fragdaspdf.de/en/other-too...
comment in response to
post
We built a tool that obfuscates those phrases to avoid detection and censorship: www.fragdaspdf.de/en/other-too...
comment in response to
post
We built a tool that obfuscates those phrases to avoid detection and censorship: www.fragdaspdf.de/en/other-too...
comment in response to
post
Dafür gibts fragdaspdf.de und im Gegensatz zu chatgpt können wir auch lange dokumente sinnvoll zusammenfassen, ohne dass Inhalt verlorengeht.
comment in response to
post
Oder fragdaspdf.de :)
comment in response to
post
Awesome work, love it
comment in response to
post
2.) Sell all your U.S. stocks
(especially those from companies that publicly support Trump’s policies)
comment in response to
post
Besser als ChatGPT bei vielen Tests glaub ich eher nicht, wir verwenden nämlich diesselben Modelle von OpenAI plus RAG
comment in response to
post
Ja fairerweise (als Jurist und Programmierer) würde auch keine KI zur Rechtsbeurteilung einsetzen, da fehlt dann das juristische Wissen im Modell
Gut funktioniert die sinner fassende Suche & Zusammenfassung im Domument
comment in response to
post
Danke fürs Sharen 😊
comment in response to
post
Inwiefern bedenklich? Wegen KI Halluzinationen meinst du?
Sollte etwas niedriger sein als bei ChatGPT und Co und man sieht die Quelle.
comment in response to
post
CC @badlogic.bsky.social