Your AI Girlfriend Is a Data-Harvesting Horror Show
The privacy mess is troubling because the chatbots actively encourage you to share details that are far more personal than in a typical app.
AI/ML
Scientists aghast at bizarre AI rat with huge genitals in peer-reviewed article
It's unclear how such egregiously bad images made it through peer-review.
llmc.sh
Adam Montgomery wrote this a neat wrapper around my LLM CLI utility: it adds a "llmc" zsh function which you can ask for shell commands (llmc 'use ripgrep to find …
GitHub - ggerganov/llama.cpp: Port of Facebook's LLaMA model in C/C++
Port of Facebook's LLaMA model in C/C++. Contribute to ggerganov/llama.cpp development by creating an account on GitHub.
Making my bookshelves clickable | James' Coffee Blog
You can make regions of an image clickable with a number of techniques, from overlyaing an SVG that contains onclick JavaScript handlers all the way to using image maps. I love this idea. I started to think about how I could create an image of my bookshelves that you could click to learn more about each book I am reading. This would be more engaging than a traditional list of text.
- Fuck You, Show Me The Prompt.
Quickly understand inscrutable LLM frameworks by intercepting API calls.
Air Canada's chatbot gave a B.C. man the wrong information. Now, the airline has to pay for the mistake
Air Canada has been ordered to compensate a B.C. man because its chatbot gave him inaccurate information.
MindMac — Elegant and Fullfil ChatGPT Client for macOS
MindMac let you use ChatGPT efficient on macOS
Function Calling in Ollama vs OpenAI
Function Calling is awesome even though it’s a terrible name for the feature. Function calling doesn't call the function, but makes it possible for you to call a function.
Sorry, posted this one before at a low resolution.
If you think I am wrong about the OpenAI feature, share an example using the core OpenAI API and let me know. I will update the video right away. A few folks have tried, but everyone points at other products from OpenAI and not the one covered in this video. Happy to admit when I am wrong, but prove it first.
Image Annotation with LLava & Ollama
LLava 1.6 models - https://huggingface.co/liuhaotian Code for this vid - https://github.com/samwit/ollama-tutorials/blob/main/ollama_python_lib/ollama_scshot...
Neural network training makes beautiful fractals
This blog is intended to be a place to share ideas and results that are too weird, incomplete, or off-topic to turn into an academic paper, but that I think may be important. Let me know what you think! Contact links to the left.
Artificial Intelligence in the News: How AI Retools, Rationalizes, and Reshapes Journalism and the Public Arena
pDownload the pdf here. Executive Summary Despite growing interest, the effects of AI on the news industry and our information environment — the public arena — remain poorly understood. Insufficient attention has also been paid to the implications of the news industry’s dependence on technology companies for AI. Drawing on 134 interviews with news workers […]/p
TfL's AI Tube Station experiment is amazing and slightly terrifying
Mind the Orwellian Surveillance Apparatus
Video: ASCII Smuggling and Hidden Prompt Instructions · Embrace The Red
ASCII Smuggling - Crafting Invisible Text and Decoding Hidden Secrets (with LLMs)
stanfordnlp/dspy at bramadams.dev
DSPy is a framework for algorithmically optimizing LM prompts and weights, especially when LMs are used one or more times within a pipeline. To use LMs to build a complex system without DSPy, you generally have to: (1) break the problem down into steps, (2) prompt your LM well until each step works well in isolation, (3) tweak the steps to work well together, (4) generate synthetic examples to tune each step, and (5) use these examples to finetune smaller LMs to cut costs. Currently, this is hard and messy: every time you change your pipeline, your LM, or your data, all prompts (or finetuning steps) may need to change.
Piping from rg to llm to answer questions about code
Here's a trick I've used a couple of times in the past few days.
How Walmart, Delta, Chevron and Starbucks are using AI to monitor employee messages
Aware uses AI to analyze companies' employee messages across Slack, Microsoft Teams, Zoom and other communications services.
Bard becomes Gemini: Try Ultra 1.0 and a new mobile app today
Bard is now known as Gemini, with a mobile app and an Advanced experience that gives you access to our most capable AI model, Ultra 1.0.
New localllm lets you develop gen AI apps locally, without GPUs | Google Cloud Blog
Want to use open-source LLM models from Hugging Face on your local development environment? With localllm and Cloud Workstations, you can.
ASCII Smuggler Tool: Crafting Invisible Text and Decoding Hidden Codes · Embrace The Red
An adversary can hide text in plain sight using the Unicode Tags. Using ASCII Smuggler you can encode and deocde such hidden messages
Hidden Prompt Injections with Anthropic Claude · Embrace The Red
Hidden Prompt Injections with Anthropic Claude
GitHub - apple/ml-mgie
Contribute to apple/ml-mgie development by creating an account on GitHub.
Apple releases ‘MGIE’, a revolutionary AI model for instruction-based image editing
Apple’s MGIE is a revolutionary AI model that can edit images based on natural language instructions, using multimodal large language models to generate expressive and imaginative edits.
Meet ‘Smaug-72B’: The new king of open-source AI
Abacus AI has released "Smaug-72B," a new open-source AI model that outperforms GPT-3.5 and Mistral Medium on the Hugging Face Open LLM leaderboard.
Vector search, RAG, and Azure AI search
I recorded myself giving the talk I gave at the SF AI tour.
Slides:
https://speakerdeck.com/pamelafox/vector-search-and-retrieval-for-generative-ai-app-microsoft-ai-tour-sf
Demos:
https://github.com/pamelafox/vector-search-demos/
https://pamelafox.github.io/vectors-comparison/
https://github.com/Azure-Samples/azure-search-openai-demo/
WebVoyager
WebVoyager: Building an End-to-End Web Agent with Large Multimodal ModelsWebVoyager is a new vision-powered web-browsing agent that uses browser screenshots ...
Beyond Self-Attention: How a Small Language Model Predicts the Next Token | Shyam's Blog
A deep dive into the internals of a small transformer model to learn how it turns self-attention calculations into accurate predictions for the next token.
‘The situation has become appalling’: fake scientific papers push research credibility to crisis point
Last year, 10,000 sham papers had to be retracted by academic journals, but experts think this is just the tip of the iceberg
cfahlgren1/natural-sql: A series of top performing Text to SQL LLMs
A series of top performing Text to SQL LLMs. Contribute to cfahlgren1/natural-sql development by creating an account on GitHub.
Vesuvius Challenge 2023 Grand Prize awarded: we can read the scrolls!
The 2000-year-old scroll discusses music, food, and how to enjoy life’s pleasures.