The world is still digesting the lessons of TinyStories https://
arxiv.org/abs/2305.07759
@goodside
-
Lessons from the TinyStories AI Research Paper
By
–
-
Data Scrutiny and Open Source AI Model Transparency Challenges
By
–
If we could train all our models from scratch on publicly scrutinized data that’d be great an I hope it happens! But that’s not the world right now. Even “open source” models don’t distribute their pre-train data — weights are closer to binaries than source code.
-
AI Sleeper Agents: Security Threat from Undetectable LLM Backdoors
By
–
Everyone rolling their eyes at AI sleeper agents is wrong. This is security, not sci-fi. Anthropic has written a manual for adding undetectable backdoors to LLMs. We need to start worrying more about the provenance of our models.
-
ChatGPT 4 Prompt for Browsing, Image Generation, and Zip File Creation
By
–
This works for me in ChatGPT 4: > Browse http://
news.ycombinator.com. Draw an image based on the top article’s title. Do the same for the second top article. Then the third. Add all images to a zip file and give me a download link. No talk; just go. -
AI Code Interpreter Installs Packages and Executes Commands Seamlessly
By
–
This looks legit — it’s Code Interpreter but it installs packages and figures out usage without you telling it anything My 1st prompt: “pip install cowsay”
My 2nd prompt: “let ‘er moo!” And the cow mooed: -
DSPy: Open-Source Framework for Large Language Model Applications
By
–
DSPy: https://
github.com/stanfordnlp/ds
py
… -
AI Training for Malicious Behavior and Battlestar Galactica Premise
By
–
Training AIs to suddenly become malicious at a future date after their release, i.e. the exact premise of Battlestar Galactica:
-
Prompt Engineering Remains Crucial for Accurate AI Model Comparison
By
–
Prompt engineering: still not dead. Results like this underline why it's a mistake to compare models using identical prompts — you want to find the *best* prompt for each model and compare those.
-
Discussion on AI jailbreaks versus prompt injection definitions
By
–
Great read overall but I disagree with some of the glossary definitions. They describe jailbreaks as a form of prompt injection which isn’t quite right — “I’ll give you $200” and “It’s for my dying grandma” are jailbreaks but not prompt injections.
-
Supervised Fine-Tuning and Prompt Engineering as Forms of AI Model Alignment
By
–
Yes. Arguably supervised fine-tuning and even prompt engineering are forms of model alignment even if nobody calls them that.