Ad
Skip to content
Subscribe Now
Matthias Bastian
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Dec 25, 2025
Short News
Copy the url to clipboard Share this article Go to comment section
Read full article about: Waymo's leaked system prompt reveals a 1,200-line rulebook for its in-car Gemini assistant

Prompt engineers, take note: Jane Manchun Wong has uncovered the system prompt for Waymo's unreleased Gemini AI assistant, a specification over 1,200 lines long buried in the Waymo app's code.

The assistant (still) runs on Gemini 2.5 Flash and helps passengers during their ride. It can answer questions, adjust the air conditioning, and change the music, but it can't steer the vehicle or alter the route. The instructions draw a clear line between the AI assistant (Gemini) and the autonomous driving system (Waymo Driver).

Waymo's system prompt follows a trigger-instruction-response pattern: a trigger defines the situation, the instruction specifies the desired behavior, and examples show wrong and correct answers. | Image: Jane Manchun Wong

The prompt uses a trigger-instruction-response pattern throughout: each rule defines a trigger, an action instruction, and often example responses. Wrong and correct answers appear side by side to clarify the desired behavior. For ambiguous questions: first clarify, then draw conclusions, finally deflect. Hard limits are enforced through prohibition lists with alternative answers. Wong's full analysis has many more details.

Comment Source: Jane Manchun Wong
Matthias Bastian
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Dec 25, 2025
Short News
Copy the url to clipboard Share this article Go to comment section
Read full article about: Salesforce executives signal declining trust in large language models

According to Salesforce leadership, confidence in large language models (LLMs) has slipped over the past year. The Information reports the company is now pivoting toward simple, rule-based automation for its Agentforce product while limiting generative AI in certain use cases.

"We all had more confidence in LLMs a year ago," said Sanjna Parulekar, SVP of product marketing at Salesforce. She points to the models' inherent randomness and their tendency to ignore specific instructions as primary reasons for the shift.

The company also struggles with "drift" - where AI agents lose focus when users ask distracting questions. Salesforce's own studies confirm this behavior remains a persistent challenge.

A spokesperson denied the company is backtracking on LLMs, stating they are simply being more intentional about their use. The Agentforce platform, currently on track for over $500 million in annual sales, allows users to set deterministic rules that strictly constrain the AI's capabilities.

Comment Source: The Information
Matthias Bastian
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Dec 24, 2025
Short News
Copy the url to clipboard Share this article Go to comment section
Read full article about: Report: OpenAI may embed sponsored content directly into ChatGPT responses

OpenAI's advertising plans for ChatGPT are taking shape. According to The Information, employees are discussing various ad formats for the chatbot. One option would have AI models preferentially weave sponsored content into their responses. So a question about mascara recommendations might surface a Sephora ad. Internal mockups also show ads appearing in a sidebar next to the response window.

Another approach would only show ads after users request further details. If someone asks about a trip to Barcelona and clicks on a suggestion like the Sagrada Familia, sponsored links to tour packages could appear. A spokesperson confirmed to The Information that the company is exploring how advertising might work in the product without compromising user trust.

OpenAI CEO Sam Altman has previously called AI responses shaped by advertising a dystopian future—especially if those recommendations draw on earlier, private conversations with the chatbot. Yet that appears to be precisely what OpenAI is now working on: advertising powered by ChatGPT's memory function, which could tap into personal conversation histories for targeted ads.

Comment Source: The Information
Matthias Bastian
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Dec 24, 2025
Short News
Copy the url to clipboard Share this article Go to comment section
Read full article about: Qwen updates image editing model with better character consistency

Qwen has released an improved version of its image editing model that better maintains facial identity during edits. The Chinese AI company published Qwen-Image-Edit-2511 on Hugging Face, an upgrade to the earlier Qwen-Image-Edit-2509. The biggest improvement is how the model handles people. It can now make creative changes to portraits while keeping the subject recognizable, the company claims. Group photos with multiple people also work better now.

The updated model can combine separate portrait images and edit group photos while preserving each person's (or cat's) identity. | Image: Qwen

The update also brings improvements to lighting control, camera angles, industrial product design, and geometric calculations. Qwen has baked popular community LoRAs (small add-on models) directly into the base model. The model ships under the Apache 2.0 license. A demo is available on Hugging Face, and users can test the model for free via Qwen Chat.

Comment Source: Hugging Face
Matthias Bastian
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Dec 23, 2025
Short News
Copy the url to clipboard Share this article Go to comment section
Read full article about: GitHub repository offers more than 50 customizable Claude Skills

A comprehensive collection of "Claude Skills" is now available on GitHub. These skills are customizable workflows that teach Anthropic's AI assistant Claude to perform specific tasks repeatedly and in a standardized way. The collection includes more than 50 skills across nine categories: Document Processing (Word, PDF, PowerPoint), Development Tools (Playwright, AWS, Git), Data Analysis, Business and Marketing, Communication, Creative Media, Productivity, Project Management, and Security.

Users can add skills in Claude.ai through the settings, store them in Claude Code's configuration folder, or integrate them via API. Each skill consists of a folder with a SKILL.md file. The repository is under Apache 2.0 license and accepts contributions.

Since skills are essentially just a collection of prompts in a folder, getting the most out of AI means customizing these prompts to fit your needs. That said, the repository is a solid source of inspiration. And as skills seem to be becoming the standard approach, it's worth exploring the topic beyond Claude.ai.

Comment Source: Github
Jonathan Kemper
Jonathan Kemper View the LinkedIn Profile of Jonathan Kemper
Dec 23, 2025
Short News
Copy the url to clipboard Share this article Go to comment section
Read full article about: Zhipu AI challenges Western rivals with low-cost GLM-4.7

Zhipu AI has introduced GLM-4.7, a new model specialized in autonomous programming that uses "Preserved Thinking" to retain reasoning across long conversations. This capability works alongside the "Interleaved Thinking" feature introduced in GLM-4.5, which allows the system to pause and reflect before executing tasks. The model shows a significant performance jump over its predecessor, GLM-4.6, scoring 73.8 percent on the SWE-bench Verified test. Beyond writing code, Zhipu says GLM-4.7 excels at "vibe coding" - generating aesthetically pleasing websites and presentations. In a blog post, the company showcased several sites reportedly created from a single prompt. Benchmark comparisons show a tight race between GLM-4.7 and commercial Western models from providers like OpenAI and Anthropic. | Image: Zhipu AI

A table with benchmark results comparing the GLM-4.7 AI model with competitors; the model shows leading values in categories such as Reasoning, Code Agent and General Agent.
Benchmark comparisons show a tight race between GLM-4.7 and commercial Western models from providers like OpenAI and Anthropic.

The model is available through the Z.ai platform and OpenRouter, or as a local download on Hugging Face. It also integrates directly into coding workflows like Claude Code. Z.ai is positioning the release as a cost-effective alternative, claiming it costs just one-seventh as much as comparable models.

Comment Source: Zhipu AI
Jonathan Kemper
Jonathan Kemper View the LinkedIn Profile of Jonathan Kemper
Dec 23, 2025
Short News
Copy the url to clipboard Share this article Go to comment section
Read full article about: Alibaba's new Qwen models can clone voices from three seconds of audio

The Qwen team at Alibaba Cloud has released two new AI models that create or clone voices using text commands. The Qwen3-TTS-VD-Flash model lets users generate voices based on detailed descriptions, allowing them to precisely define characteristics like emotion and speaking tempo. For example, a user could request a "Male, middle-aged, booming baritone - hyper-energetic infomercial voice with rapid-fire delivery and exaggerated pitch rises, dripping with salesmanship." According to the manufacturer, the model outperforms the API for OpenAI's GPT-4o mini-tts, which launched earlier this spring.

The second release, Qwen3-TTS-VC-Flash, can copy voices from just three seconds of audio and reproduce them in ten languages. Qwen claims the model achieves a lower error rate than competitors like Elevenlabs or MiniMax. The AI is also capable of processing complex texts, imitating animal sounds, and extracting voices from recordings. Both models are accessible via the Alibaba Cloud API. You can try demos for the design model and the clone model on Hugging Face.

Comment Source: Qwen