Diverse neue Prompts in mehreren Dateien
This commit is contained in:
@@ -76,3 +76,7 @@ For the deep research, my plans for the inference stack are not relevant. Any st
|
||||
I'm also okay with *multi‑GPU BF16 numbers* that illustrate the “ceiling” of un‑quantized performance.
|
||||
|
||||
---
|
||||
|
||||
Please explain in detail the differences between the 4-bit quantized GGUF model variants UD-Q4_K_XL, Q4_K_M, IQ4_XS, Q4_K_S, IQ4_NL, Q4_0, Q4_1 which I see on the Hugging Face page [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B) and also explain what GPTQ‑Int4, AWQ‑Int4 and NVFP4 means.
|
||||
|
||||
---
|
||||
|
||||
1251
Akquise.md
Normal file
1251
Akquise.md
Normal file
File diff suppressed because it is too large
Load Diff
@@ -348,3 +348,282 @@ Die Automodelle sind:
|
||||
|
||||
Der Roboter wirkt sympathisch, freundlich und akrobatisch.
|
||||
|
||||
---
|
||||
|
||||
# LinkedIn Banner füllen
|
||||
|
||||
## Motivation
|
||||
|
||||
Attached is a screenshot of my LinkedIn Banner. I want to have an art pattern on the free areas around my photo, claim text and logo.
|
||||
|
||||
## Art pattern
|
||||
|
||||
The art pattern must not touch photo, claim text or logo.
|
||||
|
||||
### Left third
|
||||
|
||||
On the left third: a stylized artistic neural network — elegant abstract nodes and refined flowing connections in soft blue (hex #3C68AE), with subtle variations in line weight. Not generic AI glow, not hexagons — organic and refined, like a thoughtful editorial illustration of neural processing.
|
||||
|
||||
### Right third
|
||||
|
||||
On the right third: a stylized electronic circuit board with flowing copper traces and subtle SMD component silhouettes (ICs, transistors, resistors, capacitors abstracted to elegant geometric forms) in soft blue (hex #3C68AE). Refined and elegant, not photo-realistic, not cluttered.
|
||||
|
||||
### Central zone
|
||||
|
||||
In the central zone: a bright, calm interweaving — circuit traces gently transitioning into neural connections, abstract chip silhouettes giving rise to network nodes. The two worlds meet without dominating each other.
|
||||
|
||||
Both worlds rendered in the same artistic language: same line weight, same level of abstraction, same color palette. Equal visual weight on both sides.
|
||||
|
||||
## Dos and Don'ts
|
||||
|
||||
No text, no logos, no human figures, no specific brand names, no recognizable products. Vector-illustration feel with subtle texture, not photographic.
|
||||
|
||||
## Colors
|
||||
|
||||
Color palette: soft blue (hex #3C68AE) as the dominant color, light gray as secondary, occasional darker blue accents for depth. Editorial magazine quality.
|
||||
Background color remains white.
|
||||
|
||||
---
|
||||
|
||||
## Motivation
|
||||
|
||||
Attached is a screenshot of my LinkedIn banner. I want to have an art pattern on the free areas around my photo, claim text and logo.
|
||||
|
||||
### Left section
|
||||
|
||||
In the left section (left of the logo): an art pattern of a stylized artistic neural network — elegant abstract nodes and refined flowing connections in soft blue (hex #3C68AE), with fine lines and subtle variations in line weight. Not generic AI glow, not hexagons — organic and refined, like a thoughtful editorial illustration of neural processing.
|
||||
|
||||
### Right section
|
||||
|
||||
In the right section (right of the logo): an art pattern of a stylized electronic circuit board with flowing copper traces and subtle SMD component silhouettes (ICs, transistors, resistors, capacitors abstracted to elegant geometric forms) in soft blue (hex #3C68AE). Refined and elegant, with fine lines, not photo-realistic, not cluttered.
|
||||
|
||||
### Transition zones
|
||||
|
||||
In the transition zones (1. Above the claim, 2. between claim and logo, and 3. below the logo): an art pattern of a minimalist, calm interweaving connecting the two worlds — circuit traces gently transitioning into neural connections. The art pattern has connections in all three transition zones.
|
||||
|
||||
Both worlds rendered in the same artistic language: same fine line weight, same level of abstraction, same color palette. Equal visual weight on both sides.
|
||||
|
||||
## Dos and Don'ts
|
||||
|
||||
You MUST NOT change, move or re-generate the given image size or ratio, foto, claim text or logo.
|
||||
The given image edges, photo, claim text and logo are providing your restricted areas!
|
||||
|
||||
No text, no logos, no human figures, no specific brand names, no recognizable products. Vector-illustration feel with subtle texture, not photographic.
|
||||
|
||||
## Art pattern arrangement
|
||||
|
||||
The art patterns must not touch image edges, photo, claim text or logo.
|
||||
|
||||
## Colors
|
||||
|
||||
Color palette: soft blue (hex #3C68AE) as the dominant color, light gray as secondary, occasional darker blue accents for depth. Editorial magazine quality.
|
||||
Background color remains white.
|
||||
|
||||
---
|
||||
|
||||
Absolutely. Here is your original prompt, completely rewritten to incorporate the strict technical and spatial hints I provided. This improved prompt is designed to eliminate ambiguity for future AI model sessions, ensuring your source photo and text are treated as immutable, untouchable assets.
|
||||
|
||||
---
|
||||
|
||||
### Gemini's Improved Prompt - Funktionierte nicht perfekt
|
||||
|
||||
## Motivation
|
||||
|
||||
Attached is a source screenshot of my existing LinkedIn banner (image_0.png). My goal is to generate a new, comprehensive art pattern across the entirety of the canvas, but ONLY within the "free areas" not occupied by my photo, claim text, and logo.
|
||||
|
||||
This is a strict **COMPOSITION ONLY** request. I am providing image_0.png as a source mask.
|
||||
|
||||
## Critical Adherence to Constraints
|
||||
|
||||
You MUST treat the following elements from image_0.png as **IMMUTABLE LAYER OVERLAYS** that cannot be altered, moved, re-sized, or re-generated. Any change to their pixel content or position is a direct failure.
|
||||
|
||||
1. **NO-CHANGE ZONE A (Text & Logo):** The entire rectangular area containing the text "Artificial Intelligence & Electronics Engineering" and the "DesT/EngS" logo block is a strictly enforced no-change zone.
|
||||
2. **NO-CHANGE ZONE B (Photo):** The entire area containing the circular portrait image is a strictly enforced no-change zone.
|
||||
|
||||
## Art Pattern Description (What to Generate)
|
||||
|
||||
Generate a sophisticated, continuous pattern in a vector-illustration style (with subtle texture) that fills the white space, respecting the boundaries of the no-change zones and leaving a 8 px wide clean white border to the entire perimeter of image_0.png, ensuring the pattern does not touch any outer edge. The pattern should have distinct thematic zones:
|
||||
|
||||
### Left section (Left of the Logo Axis)
|
||||
|
||||
Surrounding and flowing around the photo: an art pattern of a stylized artistic neural network — elegant abstract nodes and refined flowing connections in soft blue (hex #3C68AE), with fine lines and subtle variations in line weight. Not generic AI glow, not hexagons — organic and refined, like a thoughtful editorial illustration of neural processing.
|
||||
|
||||
### Right section (Right of the Logo Axis)
|
||||
|
||||
An art pattern of a stylized electronic circuit board with flowing copper traces and subtle geometric silhouettes of SMD components (ICs, transistors, resistors, capacitors abstracted to elegant forms) in soft blue (hex #3C68AE). Refined and elegant, with fine lines, not photo-realistic, not cluttered.
|
||||
|
||||
### Transition zones
|
||||
|
||||
A calm, minimalist interweaving that connects the two worlds (neural and circuit) elegantly. The interweaving must happen in three distinct transition zones relative to the *immutable* logo and text block:
|
||||
|
||||
1. In the void space directly **above** the claim text.
|
||||
2. In the void space **between** the claim text and the logo.
|
||||
3. In the void space directly **below** the logo.
|
||||
|
||||
Both worlds rendered in the same artistic language: same fine line weight, same level of abstraction, same color palette. Equal visual weight on both sides.
|
||||
|
||||
## Dos and Don'ts
|
||||
|
||||
* **DON'T** change, move, or re-generate the source photo, claim text, or logo.
|
||||
* **DON'T** let any new art pattern touch the boundaries of the image edges, the photo, the claim text, or the logo. Leave a clear white margin.
|
||||
* **DON'T** generate any text, any other logos, human figures, or recognizable branded products.
|
||||
* **DO** provide a vector-illustration feel with a subtle texture, not photographic.
|
||||
* **DO** respect the entire original image size and aspect ratio.
|
||||
|
||||
## Colors
|
||||
|
||||
* **Color palette:** soft blue (hex #3C68AE) as the dominant color, light gray as secondary, occasional darker blue accents for depth.
|
||||
* **Aesthetic:** Editorial magazine quality.
|
||||
* **Background color:** White.
|
||||
|
||||
## Required Logical Execution Path
|
||||
|
||||
To satisfy this request correctly, the model must conceptually:
|
||||
|
||||
1. Acknowledge image_0.png as a mask.
|
||||
2. Generate the full, continuous neural/circuit pattern across the entire background canvas.
|
||||
3. Mask out (erase) the pattern from all areas corresponding to the photo, text, and logo, ensuring a clean void space.
|
||||
4. Apply the *exact*, original pixel contents of the photo, claim text, and logo back into those void spaces as a perfect overlay, retaining original quality, position, and orientation.
|
||||
|
||||
---
|
||||
|
||||
**Below lets Gemini re-generate a new banner: Huge Fail!**
|
||||
**Only usable with ChatGPT but also not perfect**
|
||||
|
||||
## Motivation
|
||||
|
||||
Attached is a source screenshot of my existing LinkedIn banner (`image_0.png`). My goal is to generate a new, comprehensive art pattern in the editable white background areas only, while preserving the existing portrait, claim text, and logo exactly.
|
||||
|
||||
This is a strict **COMPOSITION ONLY** request. Treat `image_0.png` as the source image and mask reference.
|
||||
|
||||
## Absolute Preservation Requirements
|
||||
|
||||
The following elements from `image_0.png` are **IMMUTABLE LAYER OVERLAYS**. They must not be altered, moved, resized, blurred, sharpened, recolored, regenerated, or partially redrawn.
|
||||
|
||||
Any change to their pixel content or position is a direct failure.
|
||||
|
||||
1. **NO-CHANGE ZONE A — Claim Text & Logo**
|
||||
The entire rectangular area containing the claim text
|
||||
**“Artificial Intelligence & Electronics Engineering”**
|
||||
and the complete **“DesT/EngS”** logo block is strictly protected.
|
||||
|
||||
2. **NO-CHANGE ZONE B — Portrait Photo**
|
||||
The entire circular portrait area on the left is strictly protected.
|
||||
|
||||
3. **NO-CHANGE ZONE C — Outer Canvas Border**
|
||||
Keep a completely blank white border of exactly **8 px** along all four image edges: top, bottom, left, and right.
|
||||
No new artwork may enter this 8 px border.
|
||||
No line, node, dot, texture, circuit trace, component, shadow, gradient, or decorative element may touch or cross into this border.
|
||||
|
||||
## Editable Area
|
||||
|
||||
Only the white background areas outside the protected zones may be modified.
|
||||
|
||||
The new artwork must remain inside the interior editable background field, meaning:
|
||||
|
||||
- outside the portrait photo,
|
||||
- outside the claim text and logo block,
|
||||
- outside the 8 px blank white border,
|
||||
- with a visible white clearance around the portrait, text, and logo.
|
||||
|
||||
Do **not** create a full-bleed pattern.
|
||||
Do **not** place any artwork at the image edges.
|
||||
Do **not** let the pattern touch the portrait, claim text, logo, or the 8 px outer border.
|
||||
|
||||
## Art Pattern Description
|
||||
|
||||
Generate a sophisticated, continuous pattern in a vector-illustration style with subtle texture. The pattern should fill the remaining editable white space while respecting all no-change zones and margins.
|
||||
|
||||
Use distinct thematic zones:
|
||||
|
||||
### Left Section — Neural Network Motif
|
||||
|
||||
In the editable area around and flowing near the portrait, generate a stylized artistic neural network:
|
||||
|
||||
- elegant abstract nodes,
|
||||
- refined flowing connections,
|
||||
- organic line movement,
|
||||
- subtle variation in line weight,
|
||||
- soft blue as the dominant color.
|
||||
|
||||
The neural pattern should feel thoughtful and editorial, not generic “AI glow.”
|
||||
Do not use hexagons.
|
||||
|
||||
### Right Section — Electronics Motif
|
||||
|
||||
In the editable area on the right side, generate a stylized electronic circuit-board pattern:
|
||||
|
||||
- flowing copper-trace-like paths,
|
||||
- abstract SMD component silhouettes,
|
||||
- ICs, transistors, resistors, and capacitors represented as elegant geometric forms,
|
||||
- fine blue linework,
|
||||
- clean and refined composition.
|
||||
|
||||
The circuit pattern should be elegant and not photo-realistic.
|
||||
Do not make it cluttered.
|
||||
|
||||
### Transition Zones
|
||||
|
||||
Create calm, minimalist interweaving between the neural and circuit motifs in three specific editable transition areas:
|
||||
|
||||
1. the void space directly **above** the claim text,
|
||||
2. the void space **between** the claim text and the logo,
|
||||
3. the void space directly **below** the logo.
|
||||
|
||||
The transition should connect the two visual worlds elegantly, without crossing into the protected text or logo zones.
|
||||
|
||||
Both the neural and circuit motifs must use the same artistic language:
|
||||
|
||||
- same line weight family,
|
||||
- same abstraction level,
|
||||
- same color palette,
|
||||
- balanced visual density,
|
||||
- equal visual weight on both sides.
|
||||
|
||||
## Colors and Style
|
||||
|
||||
- Dominant color: soft blue `#3C68AE`
|
||||
- Secondary color: light gray
|
||||
- Accent color: occasional darker blue for depth
|
||||
- Background: pure white
|
||||
- Style: refined vector illustration with subtle texture
|
||||
- Aesthetic: editorial magazine quality
|
||||
- Linework: thin, elegant, airy, and controlled
|
||||
|
||||
## Strict Don’ts
|
||||
|
||||
- Do **not** change, move, resize, recolor, redraw, blur, sharpen, or regenerate the portrait.
|
||||
- Do **not** change, move, resize, recolor, redraw, blur, sharpen, or regenerate the claim text.
|
||||
- Do **not** change, move, resize, recolor, redraw, blur, sharpen, or regenerate the logo.
|
||||
- Do **not** let any new artwork enter the 8 px blank white border around the image.
|
||||
- Do **not** let any new artwork touch the image edges.
|
||||
- Do **not** let any new artwork touch or overlap the portrait, claim text, or logo.
|
||||
- Do **not** generate any new text.
|
||||
- Do **not** generate any new logos.
|
||||
- Do **not** generate human figures.
|
||||
- Do **not** generate recognizable branded products.
|
||||
- Do **not** use hexagons.
|
||||
- Do **not** create a full-bleed design.
|
||||
- Do **not** create a photographic or 3D-rendered look.
|
||||
|
||||
## Required Logical Execution Path
|
||||
|
||||
To satisfy this request correctly, conceptually follow this process:
|
||||
|
||||
1. Use `image_0.png` as the base image and mask reference.
|
||||
2. Identify all immutable no-change zones:
|
||||
- portrait photo,
|
||||
- claim text,
|
||||
- logo,
|
||||
- 8 px outer canvas border.
|
||||
3. Generate the neural/circuit illustration pattern only inside the remaining editable white background areas.
|
||||
4. Erase or mask out any generated pattern from all protected zones.
|
||||
5. Ensure the 8 px outer border remains completely blank white on all four sides.
|
||||
6. Restore the original portrait, claim text, and logo as exact pixel overlays in their original positions.
|
||||
7. Preserve the original image size, aspect ratio, and composition.
|
||||
|
||||
## Final Output Goal
|
||||
|
||||
The final image should look like the original LinkedIn banner with the portrait, claim text, and logo unchanged, while the formerly blank white background areas are enriched with an elegant blue neural-network and electronics-engineering illustration.
|
||||
|
||||
The result must preserve a clean 8 px white border around the entire image and clear white space around all protected elements.
|
||||
|
||||
755
Dirty-Prompts.md
Normal file
755
Dirty-Prompts.md
Normal file
@@ -0,0 +1,755 @@
|
||||
Bias von LLMs durch Trainingsdaten: Ich habe gehört, dass ein LLM die Amerikanischen Gründungsväter als teilweise nicht mit weißer Hautfarbe beschrieben hat, um politisch korrekt zu sein. Stimmt das? Falls ja, bitte Details dazu.
|
||||
|
||||
Okay, das war also nicht aufgrund der zugrunde liegenden Trainingsdaten, sondern wegen zu aggressiver Diversity‑Heuristiken im Alignment. Ich brauche bitte ein tatsächlich aufgetretenes Beispiel für LLM-Bias wegen Trainingsdaten.
|
||||
|
||||
Bitte schreibe für meine Präsentation eine kurze, zwei bis drei Sätze lange Zusammenfassung von
|
||||
a) dem Google-Image Beispiel mit den Amerikanischen Gründervätern
|
||||
b) dem Fakt, dass auch moderne LLMs wegen der Trainingsdaten zu geschlechterspezifischen Rollenverteilungen in generierten Texten neigen
|
||||
|
||||
Moderne LLMs übernehmen geschlechterspezifische Rollenklischees aus ihren Trainingsdaten: Frauen werden in generierten Texten deutlich häufiger in niedrig bezahlten oder häuslichen Tätigkeiten dargestellt, während Männer öfter mit prestigeträchtigen Berufen und Karriere assoziiert werden. Diese Muster sind ein direkter Effekt der Verzerrungen in den zugrunde liegenden Trainingstexten.
|
||||
|
||||
Bitte liefere mir ein Beispiel-Bild, was bei dem Google-Image Beispiel mit den Amerikanischen Gründervätern generiert wurde. Oder falls dir das nicht möglich ist, einen Link zu einer Beispiel-Webseite mit einem solchen Bild.
|
||||
|
||||
|
||||
Sachverhalte können sehr unterschiedliche zeitliche Relevanz haben – LLMs berücksichtigen das aktuell nicht. Bitte liefere mir zwei Beispiele dafür:
|
||||
1. Einen veralteten Sachverhalt aus einem schnellebigen Bereich (wie z.B. Generative KI), der in LLM-Trainingsdaten verwendet wurde und zu veralteten LLM-Antworten führte
|
||||
2. Einen Sachverhalt, der immer gilt (wie z.B. physikalische Zusammenhänge), der in LLM-Trainingsdaten verwendet wurde und zu zeitlos richtigen LLM-Antworten führte
|
||||
|
||||
|
||||
|
||||
Für eine Übersichts-Folie in einem KI-Workshop brauche ich eine Zusammenstellung der wichtigsten Machine-Learning Python Libraries mit jeweils kurzen Beschreibungen. Sollte ich noch weitere Tipps zu Machine-Learning Implementierungen geben?
|
||||
|
||||
Es geht um einen KI-Workshop, der KI, Machine Learning, Deep Learning und Generative KI umfasst. Ich brauche noch Folien für zwei Themenbereiche:
|
||||
1. Python Implementierung für den Bereich Machine Learning, der noch nicht Deep Learning und Generative KI umfasst.
|
||||
2. Python Implementierung für den Bereich Generative KI.
|
||||
|
||||
Bitte erstelle mir zu beiden Bereichen
|
||||
a) Eine kurze Liste der wichtigsten Python Libraries mit kurzer Beschreibung
|
||||
b) Sehr einfache Python Implementierungsbeispiele
|
||||
|
||||
|
||||
|
||||
Sorry, "Sehr einfache Python Implementierungsbeispiele" war ein Fehler. sklearn ist gut, aber Lineare Regression ist ein viel zu triviales Beispiel. Ich denke eher an etwas Anspruchsvolleres wie Anomalieerkennung mit IsolationForest aus dem sklearn.ensemble . Bitte ein solches Beispiel mit nur dem absolut erforderlichen Implementierungs-Source-Code.
|
||||
|
||||
|
||||
|
||||
Jetzt zum Generative KI Implementierungsbeispiel: Ich brauche ein Python Beispiel für eins der aktuellen SOTA-Modelle: Gpt-5.2, Gemini 3 Pro (Preview) oder Claude Opus 4.6 . Das Beispiel soll einen Textchat mit mindestens zwei User-Prompts und zwei Assistant-Prompts haben. Die Implementierung bitte so einfach wie möglich.
|
||||
|
||||
|
||||
Dein Beispiel liefert:
|
||||
{User: Welche drei Durchbrüche machten Gemini 3 Pro möglich?
|
||||
Assistant: Es gibt aktuell **kein Modell namens "Gemini 3 Pro"**.
|
||||
|
||||
Es ist sehr wahrscheinlich, dass Sie entweder das aktuelle Spitzenmodell **Gemini 1.5 Pro** meinen oder es mit dem Konkurrenzmodell **Claude 3** (von Anthropic) verwechseln.
|
||||
|
||||
Da Sie explizit nach "Gemini" fragen, beziehe ich mich hier auf die technischen Durchbrüche, die **Gemini 1.5 Pro** (den aktuellen Standard) möglich gemacht haben. Google hat hierbei drei wesentliche technologische Sprünge vollzogen: ...}
|
||||
|
||||
Offensichtlich hast du im Beispiel vergessen, das LLM auszuwählen, und das Default-LLM ist veraltet. Bitte korrigieren.
|
||||
|
||||
|
||||
|
||||
Problem with a fresh installed Debian 12 PC with German locale: When starting VSCodium, I get a window requesting a password (attached).
|
||||
|
||||
I know from previous Debian installations that there are some known issues with the key management.
|
||||
|
||||
Instead of getting ridd of the annoying window for each user separately, I prefer to fix the misconfiguration system-wide as root.
|
||||
|
||||
Please help me
|
||||
- To fully understand the problem
|
||||
- With suggestions how to fix the problem
|
||||
|
||||
Ask me questions now to start troubleshooting.
|
||||
|
||||
1. XFCE
|
||||
2. Remote Desktop connection to the PC, then enter user name and password
|
||||
3. With the attached shell script which uses the official .deb repository
|
||||
4. Window screenshot is attached
|
||||
|
||||
Option A is what I need.
|
||||
In 1. you propose to modify PAM configuration in /etc/pam.d/common-session when
|
||||
"session optional pam_gnome_keyring.so auto_start" is not in it.
|
||||
But since the file contains
|
||||
"
|
||||
# As of pam 1.0.1-6, this file is managed by pam-auth-update by default.
|
||||
# To take advantage of this, it is recommended that you configure any
|
||||
# local modules either before or after the default block, and use
|
||||
# pam-auth-update to manage selection of other modules. See
|
||||
# pam-auth-update(8) for details.
|
||||
"
|
||||
I better should use pam-auth-update - please instruct.
|
||||
|
||||
|
||||
|
||||
|
||||
I installed Debian 12 with XFCE on a new server and want to access it via
|
||||
Remote Desktop Protocol (RDP) from a Windows PC. Establishing the connection
|
||||
works, but there are some problems caused by accessing the server via RDP;
|
||||
e.g.,
|
||||
- Debian's default security policy requires administrative authentication for color management when the session is remote
|
||||
- gnome-keyring is not available for encryption
|
||||
|
||||
I have seen many web pages describing how to setup a Debian XFCE server for RDP
|
||||
access but they just describe how to establish the connection and do not
|
||||
describe how to adjust the server setup for full functionality via RDP.
|
||||
|
||||
Please search the web for an extensive instruction. If you need more details
|
||||
ask before you start the search.
|
||||
|
||||
|
||||
After 5b) a problem occurred:
|
||||
I was logged in as user "try" via RDP and became root by using "su". After
|
||||
```
|
||||
systemctl restart xrdp-sesman
|
||||
systemctl restart xrdp
|
||||
```
|
||||
the RDP connection got lost as expected.
|
||||
The problem: When user "try" logs in via RDP or at the console, the session
|
||||
immediately is terminated.
|
||||
Maybe the system got stuck in a state where user "try" still seems to be
|
||||
logged in, so what seems to become a second session is terminated.
|
||||
Other users can login via RDP or at console.
|
||||
How can I repair this so user "try" can login again?
|
||||
|
||||
|
||||
|
||||
This is my shell script to add users in a fresh Linux system:
|
||||
'''
|
||||
#!/bin/bash
|
||||
set -euo pipefail
|
||||
|
||||
echo "[20] Users setup start"
|
||||
|
||||
for user in $OPERATOR_USERS; do
|
||||
# Add user if not existing
|
||||
if ! id "$user" &>/dev/null; then
|
||||
useradd -m -s /bin/bash "$user"
|
||||
# Lock password (admin can set it later)
|
||||
passwd -l "$user"
|
||||
# Enable linger for rootless systemd services
|
||||
loginctl enable-linger "$user"
|
||||
fi
|
||||
done
|
||||
|
||||
# Ensure admin user is sudoer
|
||||
# usermod -aG sudo "$ADMIN_USER"
|
||||
|
||||
# Set root PATH in /root/.bashrc (needed when logged in user uses su)
|
||||
printf '\nexport PATH="/usr/local/sbin:/usr/sbin:/sbin:/usr/local/bin:/usr/bin:/bin"\n' >> /root/.bashrc
|
||||
|
||||
echo "User passwords need to be set manually by the administrator."
|
||||
|
||||
echo "[20] Users setup completed"
|
||||
'''
|
||||
|
||||
Now it turns out that each user needs a symbolic link from ~/Schreibtisch to ~/Desktop; e.g., for user "tlg" the command would be
|
||||
'''
|
||||
ln -s /home/tlg/Schreibtisch /home/tlg/Desktop
|
||||
'''
|
||||
|
||||
How to update the shell script?
|
||||
|
||||
|
||||
|
||||
Wie bemerkt man bei Einzelgesprächen, dass jemand zu Scientology gehört und versucht, sich mit den Regeln von Scientology einen Vorteil zu verschaffen, der einem selbst Schaden zufügen wird?
|
||||
|
||||
Vor langer Zeit hieß es, wenn man einen Anhänger von Scientology fragt, ob er den Lehren von L. Ron Hubbard folgt, dann muss er das bejahen. Stimmt das? Gibt es tatsächlich eine einfache Frage, mit der man Scientology-Anhänger entlarven kann?
|
||||
|
||||
|
||||
Anbei Screenshots meiner Anrufliste mit Telefonaten von Klaus Düsel in den Monaten Januar und Februar 2026.
|
||||
|
||||
Bitte erstelle daraus eine Liste im .csv-Format mit den Spalten
|
||||
- Zeitpunkt des Starts im Format YYYY-MM-DD_HH:MM
|
||||
- Dauer im Format HH:MM:SS
|
||||
- Richtung (eingehend, ausgehend, verpasst)
|
||||
|
||||
# Kalender-Diagramm
|
||||
|
||||
Bitte erstelle ein Kalender-Diagramm als Vektorgrafik im .svg Format,
|
||||
in dem alle Tage der Kalenderwochen
|
||||
KW4 bis KW9 des Jahres 2026 abgebildet sind und voneinander zu
|
||||
unterscheiden sind.
|
||||
|
||||
## Format
|
||||
|
||||
Wähle die optimale Einteilung und Anordnung für ein Gesamtformat
|
||||
von 16:9 . Die einzelnen Tage brauchen nicht beschriftet zu sein,
|
||||
jede Kalenderwoche soll die Zahl der Kalenderwoche in hellgrau
|
||||
im Hintergrund haben.
|
||||
|
||||
## Kalendereinträge
|
||||
|
||||
Im Kalender sind zwei verschiedene Listen von Zeitabschnitten
|
||||
eingezeichnet.
|
||||
|
||||
### Kalendereinträge aus Liste 1
|
||||
|
||||
Liste 1 ist zu erstellen, indem aus dem angehängten
|
||||
Time_Tracker_Januar_Februar_2026.ics
|
||||
alle "lumiz" Einträge extrahiert werden.
|
||||
|
||||
Die Einträge aus Liste 1 sollen in blauer Farbe in den Kalender
|
||||
eingetragen werden; das sind die Arbeitszeiten.
|
||||
|
||||
### Kalendereinträge aus Liste 2
|
||||
|
||||
Liste 2 ist zu erstellen, indem aus dem angehängten
|
||||
Düsel-Telefonate.csv
|
||||
alle "lumiz" Einträge extrahiert werden.
|
||||
|
||||
Die Einträge aus Liste 2 sollen in roter Farbe in den Kalender
|
||||
eingetragen werden; das sind die Zeiten der Telefonate.
|
||||
|
||||
|
||||
---
|
||||
|
||||
I read that prompts work better when they are formatted.
|
||||
Possible formattings are Markdown, XML and JSON.
|
||||
I usually use Markdown for other LLMs.
|
||||
Since I also read that Anthropic LLMs work best with prompts formatted with XML I would like to ask you: Which is the best formatting for Opus 4.6, Sonnet 4.6 and Haiku 4.5?
|
||||
|
||||
---
|
||||
|
||||
Der Muskel-Manager
|
||||
Aus Sjards YouTube Weisheiten
|
||||
Sjard Roscher YouTube Videos als Input, Claude für KI Buchtexte, Gemini für KI Bilder
|
||||
|
||||
---
|
||||
|
||||
Bitte sieh dir die angehängte claude.md an, die damit beschriebene App möchte ich erweitern.
|
||||
|
||||
Ich würde gern Step 4 Generierung in "Text Generierung" umbenennen.
|
||||
|
||||
Bei Step 5 soll dann für jedes Kapitel der Fließtext nicht nur wie jetzt in einem Unter-Fenster angezeigt werden, sondern es sollen diese Erweiterungen und Änderung eingebaut werden:
|
||||
- Step 5 soll "Image Generierung" heißen.
|
||||
- Die App soll durch KI ermitteln, an welchen Stellen der Text durch Images ergänzt werden sollte. An diesen Stellen soll zwischen den Fließtext-Zeilen eine Box oder ein weiteres Unter-Fenster eingebaut werden ("Add Image Box"), in der beschrieben wird, welches Image eingefügt werden könnte und warum.
|
||||
- Der User soll
|
||||
- In einer Add Image Box die Beschreibung des Images und die Begründung für das Image editieren können,
|
||||
- Die Add Image Boxen an andere Stellen des Fließtextes verschieben können,
|
||||
- Das durch eine Add Image Box vorgeschlagene Image akzeptieren oder ablehnen können,
|
||||
- Neue Add Image Boxen im Fließtext einbauen können,
|
||||
- Alle Add Image Boxen mit einem Klick akzeptieren können,
|
||||
- Alle Add Image Boxen mit einem Klick ablehnen können.
|
||||
- Sobald alle Add Image Boxen akzeptiert oder abgelehnt wurden, sollen die Images durch KI erzeugt und unter den Add Image Boxen angezeigt werden. Nachdem das erste Image erstellt wurde, soll der User
|
||||
- Bei jedem Image den dazu gehörenden Add Image Box Text editieren und das Image erneut erstellen lassen können,
|
||||
- Jedes Image entfernen können,
|
||||
- Jedes Image akzeptieren können,
|
||||
- Alle Images mit einem Klick akzeptieren können,
|
||||
- Alle Images mit einem Klick ablehnen können.
|
||||
- Die Download-Möglichkeit soll aus Step 5 entfernt werden (soll im neuen Step 6 sein).
|
||||
|
||||
Ein neuer Step 6 "Download" soll hinzugefügt werden, der, wie beim bisherigen Step 5, für jedes Kapitel eine Vorschau zeigt und der das fertige Buch zum Download bereit stellt.
|
||||
|
||||
Bitte bewerte meine Ideen und lass uns bei einem Brainstorming die Details oder Optimierungen klären.
|
||||
|
||||
|
||||
Dankeschön. Beim Testen haben sich noch zwei Änderungswünsche ergeben:
|
||||
|
||||
1. Es ist doch besser, abgelehnte Boxen im Editor-Unterfenster zu lassen. So, wie die akzeptierten Boxen grün gefärbt werden und ein Häkchen mit "Akzeptiert" bekommen, sollen die abgelehnten Boxen rot gefärbt werden und ein X mit "Abgelehnt" bekommen. Bitte ändern.
|
||||
|
||||
2. Nach jedem akzeptieren oder ablehnen einer Box schließt sich das Editor-Unterfenster. Das ist inakzeptabel, weil der User das Editor-Unterfenster wieder öffnen muss und an die aktuelle Stelle fürs Editieren scrollen muss. Bitte verbessern: Das Editor-Unterfenster soll nach akzeptieren oder ablehnen einer Box geöffnet und an der gleichen Position bleiben.
|
||||
|
||||
|
||||
Danke, aber das ist so noch nicht brauchbar, bitte überarbeiten. Diese Probleme haben meine Tests gezeigt:
|
||||
|
||||
1. Die Version wurde nicht hochgezählt. Die Version, die du jetzt erstellen wirst, ist 1.3.2 .
|
||||
|
||||
2. Bei Step 3 steht auf dem Button "Buch generieren", bitte die Beschriftung ändern auf "Text generieren".
|
||||
|
||||
3. In Step 5 Image-Generierung sieht der Button "Weiter zum Download" aktiv aus, wenn die Bilder noch nicht generiert sind. Bitte ändern: Der Button "Weiter zum Download" soll so lange deaktiviert sein, wie die Bilder noch nicht generiert sind. Erst wenn alle Bilder generiert sind, soll der User auf "Weiter zum Download" klicken können.
|
||||
|
||||
4. Die Generierung der Bilder funktioniert nicht. In den Boxen erscheint "Fehler bei Generierung" und daneben ein Button "Erneut versuchen". Auch bei erneut versuchen funktioniert die Generierung nicht.
|
||||
|
||||
---
|
||||
|
||||
# AI Agents Environment
|
||||
|
||||
## Motivation
|
||||
|
||||
For SW design in a structured flow I want to use multiple AI agents
|
||||
which are orchestrated automatically in an FOSS AI agents environment
|
||||
running on-premise on my server or PC.
|
||||
|
||||
## Candidates
|
||||
|
||||
Popular candidates are:
|
||||
1. Superpowers, https://github.com/obra/superpowers
|
||||
2. Everything Claude Code, https://github.com/affaan-m/everything-claude-code
|
||||
|
||||
## First impressions
|
||||
|
||||
### Superpowers
|
||||
|
||||
Seems to be a good match because of the Seven-step Basic Workflow.
|
||||
Looks like a persistent memory is missing.
|
||||
|
||||
### Everything Claude Code
|
||||
|
||||
Looks like a certain level of expert knowledge is required because
|
||||
no structured flow is provided.
|
||||
|
||||
## Your tasks
|
||||
|
||||
### Interview
|
||||
|
||||
First interview me to understand my preferences, capabilities and
|
||||
requirements.
|
||||
|
||||
### Comparison
|
||||
|
||||
- Are other AI Agent environments possible candidates?
|
||||
- Compare the candidates in a comprehensive report including
|
||||
Pros and Cons
|
||||
- Recommend the candidate which is best suited for my purposes
|
||||
|
||||
|
||||
|
||||
I'd be happy to help you find the best AI agents environment for your needs. Let me start by understanding your situation better before diving into research.
|
||||
|
||||
Here are a few open-ended questions first:
|
||||
|
||||
**What kind of SW design work are you doing?** For example, are you building web apps, embedded systems, microservices, data pipelines — and roughly how large are these projects?
|
||||
|
||||
**What does "structured flow" mean to you concretely?** For instance, are you thinking of something like: requirements → architecture → implementation → testing → review? Or something more granular?
|
||||
|
||||
Now let me gather a few more specifics:Take your time with these — your answers (along with the open-ended questions above) will help me tailor the comparison and recommendation precisely to your situation.
|
||||
|
||||
|
||||
|
||||
I have two setups:
|
||||
1. Windows 10 notebook
|
||||
2. Debian Linux version 12 server with NVIDIA RTX 5070 Ti 16 GB VRAM
|
||||
|
||||
|
||||
|
||||
Excellent work, thank you! For coding I will use Superpowers and
|
||||
accept being limited to Claude LLMs.
|
||||
|
||||
Now please advice for another kind of implementation and test
|
||||
which is more infrastructure setup and less SW design:
|
||||
Selecting (with human-in-the-loop)
|
||||
several FOSS SW packages which can run as
|
||||
microservices, write shell scripts to create and run
|
||||
Podman Pods with those SW packages, design some small
|
||||
additional SW packages like MCP servers or proxies, test
|
||||
the services and fix bugs.
|
||||
|
||||
Which AI Agent environment would you recommend for this and why?
|
||||
|
||||
|
||||
Superpowers can also run as a plugin for OpenCode.
|
||||
I do not know OpenCode, please explain what it is, how it compares
|
||||
against Claude Code and which
|
||||
LLMs can be used in OpenCode. Maybe it is even possible
|
||||
to use my on-premise LLMs with it?
|
||||
|
||||
|
||||
|
||||
No, but please create an additional comprehensive report about our
|
||||
conversation about the infrastructure-focused work, the possibilities
|
||||
to switch off / work without the Superpowers plugin and OpenCode as
|
||||
the Claude Code alternative which also allows on-premise LLM usage.
|
||||
Provide this report again in raw Markdown format for download.
|
||||
|
||||
|
||||
---
|
||||
|
||||
# FOSS On-Premise Web-GUI drawing canvas tool
|
||||
|
||||
A FOSS On-Premise Web-GUI drawing canvas tool is needed for
|
||||
professional visual planning purposes.
|
||||
|
||||
Two tools were evaluated but are not well suited.
|
||||
|
||||
Finding a ready-to-use tool which satisfies all needs is preferred
|
||||
over building an own canvas tool with FOSS libraries and self-written
|
||||
software.
|
||||
|
||||
## Tools evaluated up to now
|
||||
|
||||
### AFFiNE
|
||||
|
||||
AFFiNE was evaluated but did not show the expected performance
|
||||
when doing copy-paste of shape groups.
|
||||
AFFiNE is not completely FOSS.
|
||||
AFFiNE has no MCP server and not an official API.
|
||||
Despite of these points the Web-GUI drawing canvas look-and-feel is acceptable.
|
||||
|
||||
### Miro
|
||||
|
||||
Miro is not FOSS.
|
||||
The Web-GUI drawing canvas look-and-feel is acceptable.
|
||||
|
||||
## Libraries for a self-made tool
|
||||
|
||||
Several FOSS canvas libraries allow to design an own On-Premise Web-GUI
|
||||
drawing canvas tool.
|
||||
|
||||
### Candidates
|
||||
|
||||
Candidates are
|
||||
- Fabric.js
|
||||
- Konva.js
|
||||
|
||||
### Likely not matching
|
||||
|
||||
As far as known up to now, these are not matching the needs:
|
||||
- Excalidraw
|
||||
- React Flow
|
||||
|
||||
## Your task
|
||||
|
||||
- Search the net for ready-to-use Web-GUI drawing canvas tools based
|
||||
on Fabric.js or Konva.js
|
||||
|
||||
---
|
||||
|
||||
# Buchhaltung der Rechnungen nach einem Vergleich
|
||||
|
||||
Anbei die Beschreibung einer Sachlage.
|
||||
|
||||
Frage: Wie gehe ich am besten damit in meiner Buchhaltung um?
|
||||
|
||||
Ich nutze ein Buchhaltungs-Programm und den Kontenrahmen SKR-03.
|
||||
Die ursprüngliche Rechnung AR02002 habe ich noch nicht gebucht.
|
||||
|
||||
## Idee 1
|
||||
|
||||
Ich ersetze die ursprüngliche Rechnung AR02002 durch eine reduzierte
|
||||
Rechnung AR02002a mit dem Betrag 4.760,00 € und buche diese.
|
||||
|
||||
Favorisiert, weil am wenigsten buchhalterischer Arbeitsaufwand.
|
||||
|
||||
## Idee 2
|
||||
|
||||
Ich buche für Februar 2026
|
||||
die ursprüngliche Rechnung AR02002 mit dem Betrag 13.090,00 €
|
||||
und erzeuge einen offenen Posten auf einem Lumiz Debitorenkonto.
|
||||
|
||||
Im März bilde ich dann den Vergleich ab. Wie genau ich das machen
|
||||
würde, weiß ich noch nicht. Wahrscheinlich den entgangenen Betrag
|
||||
auf ein spezielles Buchhaltungskonto buchen, auch eine reduzierte
|
||||
Rechnung AR02002a mit dem Betrag 4.760,00 € erstellen und buchen.
|
||||
|
||||
## Dein Job
|
||||
|
||||
Bitte bewerte meine Ideen. Gibt es noch andere Möglichkeiten?
|
||||
Was empfiehlst du?
|
||||
|
||||
|
||||
|
||||
|
||||
Danke für deine Analyse und Tipps.
|
||||
|
||||
In meiner Buchhaltung stehen die ersten beiden Ziffern einer Rechnungs-Belegnummer
|
||||
hinter "AR" für den Monat und die letzten drei Ziffern sind eine
|
||||
laufende Nummer. Die ursprüngliche Rechnung AR02002 war also im Februar
|
||||
die zweite Rechnung.
|
||||
|
||||
Ich sollte nun also:
|
||||
- Die Storno-Rechnung AR02002-S nennen und das Datum von heute verwenden,
|
||||
damit man sie der ursprünglichen Rechnung zuordnen kann.
|
||||
- Die neue Rechnung AR03001 nennen und ebenfalls
|
||||
das Datum von heute verwenden.
|
||||
|
||||
Korrekt?
|
||||
|
||||
|
||||
|
||||
|
||||
Ich möchte das Geschehen auch in der Buchhaltung vollständig abbilden
|
||||
und an jedem Monatsende alle Vorfälle des Monats dokumentiert haben.
|
||||
Daher möchte ich für Februar 2026
|
||||
die ursprüngliche Rechnung AR02002 mit dem Betrag 13.090,00 € buchen
|
||||
und erzeuge dadurch einen offenen Posten auf einem Lumiz Debitorenkonto.
|
||||
Bei der Buchhaltung für März wird dieser offene Posten durch die
|
||||
Storno-Rechnung AR02002-S dann auf Null gesetzt und die neue Rechnung
|
||||
AR03001 kommt zum Tragen.
|
||||
|
||||
Ja, bitte hilf mir, den rechtssicheren Text für die Storno-Rechnung zu formulieren.
|
||||
|
||||
|
||||
---
|
||||
|
||||
# Company-wide item numbering system
|
||||
|
||||
## Idea
|
||||
Idea: Have a company-wide item numbering system using sha256 format.
|
||||
Run a microservice which provides a database in which to each of
|
||||
the items (with the item number as key) the key facts of the item
|
||||
are stored.
|
||||
|
||||
## Display style
|
||||
Usually sha256 numbers are displayed for humans as a string with
|
||||
hex numbers which
|
||||
results in 64 digits in the range 0 to F, 16 in total.
|
||||
This has two drawbacks:
|
||||
1. Because only 16 different digits are in use, the string is longer
|
||||
than necessary.
|
||||
2. A human reader can easily mix up certain digits, e.g. 'I' with 'l' or '1';
|
||||
'O' with '0'; 'B' with '8'.
|
||||
|
||||
### Bitcoin format
|
||||
For Bitcoins there is a format designed to prevent that humans
|
||||
can mix up certain digits - please inform about the details.
|
||||
|
||||
### Alternative formats
|
||||
Please show which alternative formats were introduced which improve
|
||||
displaying a sha256 number for humans regarding
|
||||
- Shortest possible string
|
||||
- Prevention of mixing digits up
|
||||
|
||||
## Existing FOSS solutions
|
||||
|
||||
Are there any Free and Open Source Software solutions available
|
||||
which allow usage for commercial purpose without costs?
|
||||
|
||||
|
||||
|
||||
My idea was to use the 32 bytes of the sha256 number as the database
|
||||
key and just use the Base58 number wherever the number is displayed
|
||||
to humans.
|
||||
|
||||
Your Architecture Sketch shows the Base58 number as key.
|
||||
|
||||
Is there an issue with the sha256 number as the database
|
||||
key?
|
||||
|
||||
|
||||
|
||||
Back to the display challenge: It happened several times that
|
||||
I mixed up 'B' and '8'. Unfortunately Base58 does have those
|
||||
two digits so in my judgement it is not Confusion-Safe.
|
||||
Are there alternatives which focus on being Confusion-Safe which
|
||||
prevent the three confusion risks
|
||||
'I' with 'l' or '1'; 'O' with '0'; 'B' with '8'?
|
||||
|
||||
Regarding Crockford Base32:
|
||||
- Is it preventing the three confusion risks?
|
||||
- Are the hyphens optional?
|
||||
|
||||
|
||||
If we would replace the 'B' in Base58 by another character to
|
||||
make it more Confusion-Safe - which characters could be candidates?
|
||||
Could be special characters which won't cause trouble (e.g., in
|
||||
links or filenames) and are not easy to mix up by humans.
|
||||
|
||||
---
|
||||
|
||||
# Open WebUI container version update
|
||||
|
||||
## Motivation
|
||||
I'm running Open WebUI in a container, currently with this image:
|
||||
'ghcr.io/open-webui/open-webui:git-9ae06a3'.
|
||||
It is used in a professional environment so it is mandatory to use
|
||||
a stable, tagged image version.
|
||||
My current Open WebUI container image is old and I need to update it.
|
||||
|
||||
## New image
|
||||
I use local AI hardware with CUDA.
|
||||
According to the [Open WebUI Documentation](https://docs.openwebui.com/):
|
||||
- "For production environments where stability and reproducibility
|
||||
are important, it is recommended to pin a specific release version".
|
||||
- Example for pinned version with CUDA support (version number just
|
||||
for illustration purpose):
|
||||
"ghcr.io/open-webui/open-webui:v0.8.6-cuda".
|
||||
|
||||
## Your task
|
||||
Please determine the latest pinned Web UI container version for
|
||||
production environments.
|
||||
|
||||
---
|
||||
|
||||
There are AI models on Hugging Face which were tuned to remove the
|
||||
original model's refusals; they usually have "uncensored" or "NSFW"
|
||||
in the name.
|
||||
|
||||
I'm not in the details regarding this area but I want to evaluate
|
||||
such models.
|
||||
|
||||
My GPU has 16 GB VRAM; therefore, I focus on these original models:
|
||||
- [Hugging Face lovedheart/Qwen3.5-9B-FP8](https://huggingface.co/lovedheart/Qwen3.5-9B-FP8)
|
||||
- [Hugging openai/gpt-oss-20b](https://huggingface.co/openai/gpt-oss-20b)
|
||||
|
||||
Please search for uncensored models which are derived from my
|
||||
original models and recommend one.
|
||||
I want to have as many refusals removed as possible but the model's
|
||||
remaining capabilities should be as original, especially the intelligence.
|
||||
|
||||
---
|
||||
|
||||
# GPU monitoring tool
|
||||
|
||||
Hello, please help me to find a GPU monitoring tool for my AI server
|
||||
which satisfies the requirements shown below.
|
||||
|
||||
## Requirements
|
||||
|
||||
### OS
|
||||
|
||||
Must run under Linux Debian version 12 with XFCE desktop.
|
||||
|
||||
### GPU
|
||||
|
||||
Must be able to monitor a NVIDIA RTX 5070 Ti GPU.
|
||||
|
||||
### Monitoring
|
||||
|
||||
Must provide realtime history graphs of at least
|
||||
- VRAM usage
|
||||
- Power dissipation
|
||||
|
||||
## Rejected candidates
|
||||
|
||||
Another AI recommended tools which were not acceptable:
|
||||
|
||||
### LACT
|
||||
|
||||
LACT did not find the NVIDIA RTX 5070 Ti.
|
||||
|
||||
### Nvitop
|
||||
|
||||
Nvitop does not show the GPU power dissipation versus time in a
|
||||
history graph.
|
||||
|
||||
---
|
||||
|
||||
What are the best ways to let AI Agents run tests which include monitoring the dynamic output of web pages? Example use-cases are:
|
||||
|
||||
## Open WebUI
|
||||
|
||||
A design project which uses Open WebUI has severe problems but the AI Agent coding harness is not yet checking Open WebUI's web page outputs. An error
|
||||
message which appears after a second and then disappears after a few seconds would have to be detected by the testing AI Agent.
|
||||
|
||||
## UI.Vision
|
||||
|
||||
When automating a webshop buying process, many webpage elements were not trivial to access. A testing AI Agent would need to watch whether the programmed action successfully executes or wether the webpage element needs to be accessed with another method. The UI.Vision program executes actions on the webpage via a UI.Vison browser plugin.
|
||||
|
||||
---
|
||||
|
||||
# Promotion im Lebenslauf besonders beschreiben
|
||||
|
||||
## Motivation
|
||||
|
||||
Ich überarbeite gerade meinen Lebenslauf, dabei möchte ich hervorheben, dass ich für meine Promotion (Dr.-Ing.) deutlich mehr gearbeitet habe als üblich.
|
||||
|
||||
## Arbeitsumstände
|
||||
|
||||
Im Normalfall bekommen Ingenieure den Doktortitel durch eine Doktorandenstelle: Entweder während einer Tätigkeit an der Uni oder wenn sie als Wissenschaftler an einem Forschungsinstitut arbeiten. In beiden Fällen wird ihnen für die Arbeit an der Promotion ein Teil ihrer Arbeitszeit dafür bewilligt.
|
||||
|
||||
Ich hatte als Wissenschaftler an einem Forschungsinstitut eine Vollzeit-Stelle, bei der keine Zeit für eine Promotion vorgesehen war. Neben dieser Stelle habe ich in meiner Freizeit an einer Uni mit dem Thema der Vollzeit-Stelle eine Promotion begonnen, die ich erst abgeschlossen habe, nachdem ich schon mehrere Jahre lang die Wissenschaftler-Stelle nicht mehr hatte, sonder schon in der Industrie einen Job hatte.
|
||||
|
||||
## Formulierungen
|
||||
|
||||
Wie kann ich in meinem Lebenslauf elegant dem Leser zu verstehen geben, dass ich keine normale Doktoranden-Stelle hatte, sondern mir meinen Dr.-Titel mit erheblicher Mehrarbeit erarbeitet habe?
|
||||
|
||||
### Ganz alte Formulierung
|
||||
|
||||
"Wissenschaftlicher Mitarbeiter am Ferdinand-Braun-Institut für Höchstfrequenztechnik, Berlin:
|
||||
• Erzeugung elektrischer Transienten im Picosekundenbereich mit einer nichtlinearen Diodenleitung (Promotionsthema): ..."
|
||||
|
||||
### Etwas neuere Formulierung, noch ungeschickter
|
||||
|
||||
"Wissenschaftlicher Mitarbeiter am Ferdinand-Braun-Institut für Höchstfrequenztechnik, Berlin:
|
||||
• Promotion: Erzeugung elektrischer Transienten im Picosekundenbereich mit einer nichtlinearen Diodenleitung ..."
|
||||
|
||||
---
|
||||
|
||||
## Headset OK
|
||||
|
||||
Dankeschön, der Tipp mit PhonerLite interne Sound-Einstellungen hat das Headset wieder funktionieren lassen.
|
||||
|
||||
## Verbinden neuer Nummern funktioniert nicht
|
||||
|
||||
Leider habe ich noch ein sehr merkwürdiges, anderes Problem:
|
||||
- Anrufen bei zuvor im Telefonbuch vorhandenen Nummern funktioniert.
|
||||
- Im Logbuch steht "Normal call clearing".
|
||||
- Anrufen bei heute ins Telefonbuch eingetragenen Nummern funktioniert nicht: Nach doppelklicken oder Rechte Maustaste + "Verbinden" klicken wird die Verbindung nicht aufgebaut.
|
||||
- Im Logbuch steht "400:Could not parse request URI".
|
||||
|
||||
PhonerLite zeigt unten links den Status registriert an (grüne LED).
|
||||
|
||||
Woran kann das denn liegen?
|
||||
|
||||
---
|
||||
|
||||
Assuming, after working a while in a Claude Code session valuable results have been obtained which are scattered among the chat-conversion context.
|
||||
|
||||
Starting from this situation, I would like to proceed with different follow-up tasks. Each follow-up task will be a one-time job so there is no need for a Claude skill generation or a tool creation.
|
||||
|
||||
Please describe in general which options I have to conserve the original Claude Code session situation with the chat-conversion context so that I can use it later for follow-up tasks whenever I need.
|
||||
|
||||
It makes no sense to ask for a current situation because this is a question of general interest.
|
||||
|
||||
---
|
||||
|
||||
# Points to take a deeper look
|
||||
|
||||
My goal is to preserve a Claude Code session as it is.
|
||||
I do not want to start a process of generating something from the session because any extraction or description process implies the risk of loosing information.
|
||||
|
||||
## 1. Built-in session resume
|
||||
|
||||
I tried this before but I had the impression that when starting a follow-up task <Task-A> session with `claude --resume <session-code-123>` it works fine in this session.
|
||||
But when later starting another follow-up task <Task-B> session with `claude --resume <session-code-123>` the work of <Task-A> was added to the context of <session-code-123>. Am I wrong?
|
||||
|
||||
## 2. Export the transcript to a file
|
||||
|
||||
Bullet points 1 or 2 likely are what I want to have.
|
||||
|
||||
Is there any benefit of using the `/export` slash command when I do not need a readable content?
|
||||
|
||||
### Raw session logs
|
||||
|
||||
Please explain in detail how to handle the raw session logs, including how to provide them to a new Claude session.
|
||||
|
||||
---
|
||||
|
||||
For my NVIDIA RTX5070 Ti GPU with 16 GB VRAM I am using quantized Qwen AI inference models. The model Qwen3.5-9B with 8-bit quantization fits well into my VRAM with big margin. Unfortunately the model Qwen3.6-27B with 4-bit quantization does barely fit into my VRAM and I must limit the context window size to operate it.
|
||||
|
||||
The Qwen3.6 model series has improved intelligence compared to the Qwen3.5 model series. Up to now on Qwen3.6-9B model or any other Qwen3.6 series model with less than 27B parameters was published.
|
||||
|
||||
Please perform a deep research to find out which Open Weights Qwen models are in the pipeline of Qwen. I hope for a model which has the improved intelligence of the Qwen3.6 series and has lower size than 27B.
|
||||
|
||||
---
|
||||
|
||||
Thank you but this is not what I asked for.
|
||||
|
||||
I wrote "Please perform a deep research to find out which Open Weights Qwen models are in the pipeline of Qwen."
|
||||
|
||||
I already know which models are published and that no Open Weights Qwen3.6 with less than 27B was published.
|
||||
|
||||
Your task is to find out what Qwen has in the pipeline. That means on which Open Weights Qwen3.6 (or even newer) models Qwen currently is working without having them released yet.
|
||||
|
||||
You might want to study Qwen announcements or posts of Qwen employees to find this out.
|
||||
|
||||
---
|
||||
|
||||
Ich bin Freiberufler, es geht um die Freiberufler-Plattform malt.de und ich befürchte, dass ich bei länger dauernden Projekten über malt.de erst am Ende des Projekts meine Bezahlung bekommen würde.
|
||||
|
||||
Bitte führe einen Deep Research aus um zu ermitteln, ob die Freiberufler bei malt.de Zwischenrechnungen einreichen können und während des Projekts Zwischenzahlungen erhalten können, oder ob sie bis zum Projektende auf die Zahlungen warten müssen. Meine Projekte liefen bisher fast immer viele Monate - so lange auf eine Zahlung zu warten (und bei zahlungsunwilligen Endkunden vielleicht einen Zahlungsausfall von Monaten zu haben) kann und will ich mir nicht leisten.
|
||||
|
||||
---
|
||||
|
||||
Bitte führe einen Deep Research aus, um für die Freiberufler Plattform FRATCH [Fratch Website](https://fratch.io/de/) zu ermitteln:
|
||||
- Das Geschäftsmodell - wodurch genau verdient FRATCH sein Geld?
|
||||
- Die Kosten für Unternehmen und für Freiberufler
|
||||
- Vorteile, Zahlungsbedingungen inklusive Zahlungsfristen und Risiken für Freiberufler
|
||||
|
||||
---
|
||||
|
||||
Wir sind beim Portal Kleinanzeigen auf einen Betrüger reingefallen.
|
||||
|
||||
Wir hatten etwas zum Verkauf angeboten. Der Käufer hatte behauptet, die Paypal Überweisung würde nicht funktionieren, und er hätte deswegen einfach auf Direktkauf geklickt. ER schrieb, wir hätten bereits eine E-Mail von Kleinanzeigen bekommen.
|
||||
|
||||
Es war auch eine E-Mail da - vorgeblich von Kleinanzeigen, war auch so gestaltet. Leider war das eine Pishing E-Mail. In der E-Mail wurde meine Frau aufgefordert, die Bankverbindung einzurichten (sie ist neu bei Kleinanzeigen). Giro- und Kreditkarten Nummern (inkl. CSV) wurden angefordert, "Probe-Abbuchungen" wurden vorgenommen und meine Frau hat Fotos ihres Personalausweises mit dem Smartphone gemacht und ihr Gesicht in einem Video-Ident-Verfahren aufgenommen.
|
||||
|
||||
Das war aber alles Betrug. Die Karten hat meine Frau sperren lassen.
|
||||
|
||||
Unsere größte Sorge ist nun, dass der Betrüger mit den Fotos des Personalausweises und dem aufgenommenen Gesichts-Video die Identität meiner Frau nutzen könnte, um sich in ihrem Namen Geld zu verschaffen.
|
||||
|
||||
Stelle mir zwischen drei und fünf Fragen, um die Situation vollständig zu verstehen.
|
||||
|
||||
Dann berate uns, wie wir am besten reagieren sollten.
|
||||
|
||||
---
|
||||
|
||||
16
Finanzanlagen.md
Normal file
16
Finanzanlagen.md
Normal file
@@ -0,0 +1,16 @@
|
||||
Bitte sieh dir den Kurs des Zertifikats WKN HAFx4V an. Das ist ein Rohstoffe-Fonds, der auf seltene Erden und ähnliches setzt.
|
||||
|
||||
Seit Mitte 2022 ist der Kurs drei Jahre lang gefallen (grob von 250 € auf 100€), in der zweiten Hälfte von 2025 wieder schnell angestiegen (grob auf 250 €), seitdem hat er sich seitwärts bewegt (grob zwischen 250 und 270 € mit einem kurzen Einbruch in der zweiten März-Hälfte auf 200 €).
|
||||
|
||||
Ich denke darüber nach, meine Anteile zu verkaufen und brauche eine fundierte Analyse, warum sich der Kurs seit Mitte 2022 so entwickelt hat und wann der sinnvollste Zeitpunkt zum Verkauf ist.
|
||||
|
||||
Gibt es globale Entwicklungen oder Änderungen im Bereich Rohstoffe oder seltene Erden, die eine Kursänderung nach oben oder nach unten erwarten lassen?
|
||||
|
||||
---
|
||||
|
||||
Dankeschön.
|
||||
|
||||
Anbei ein Screenshot der Handelsplätze und ihrer Daten heute Morgen.
|
||||
|
||||
Ich möchte nun mit Limit verkaufen. Wo sollte ich meine Verkaufs-Order setzen und welches Limit sollte ich setzen? Oder empfiehlt es sich, noch ein paar Minuten zu warten?
|
||||
|
||||
@@ -62,3 +62,4 @@ Please examine these three examples of such shell scripts:
|
||||
2. Ask me before proceeding with task 3.
|
||||
3. Store the knowledge which you gained in this session to the
|
||||
appropriate Claude files.
|
||||
|
||||
|
||||
12
Podcast-Folgen.md
Normal file
12
Podcast-Folgen.md
Normal file
@@ -0,0 +1,12 @@
|
||||
# Hat nicht funktioniert:
|
||||
|
||||
Ich bin TÜV-zertifizierter AI Consultant und promovierter Ingenieur mit über 30 Jahren Erfahrung in Elektronik Entwicklung, Test und System Integration.
|
||||
|
||||
Für ca. 4 Stunden Autofahrt brauche ich bitte Vorschläge für Podcast-Folgen über KI, die mich nicht langweilen. Achtung: Es reicht nicht, Podcasts vorzuschlagen, weil ich nur ausgewählte Folgen brauche, die meinen Interessen entsprechen.
|
||||
|
||||
Aktuell arbeite ich meist mit Claude Code (auch mit der Extension Superpowers) und Claude Cowork. Ich habe vor, Open Claw zu nutzen, habe damit aber noch nicht begonnen. Das Thema professionelle On-Premise KI interessiert mich, wobei ich von den Grenzen meiner RTX 5070 Ti ernüchtert bin und mehr Wissen über GPUs oberhalb der Consumer Grafikkarten gebrauchen könnte.
|
||||
|
||||
Ich habe Spotify.
|
||||
|
||||
Bitte schlage mir passende Folgen vor.
|
||||
|
||||
Reference in New Issue
Block a user