OpenAI disables internet access during model training while Anthropic issues another mea culpa
Published
What's this about?
We've totally got things under control
The one-panel cartoon
Read the cartoon's script
A soundproof recording studio with a large mixing console and speakers, where an engineer wearing noise-cancelling headphones is trying to train a machine learning model with the internet disconnected, while a voiceover in the background repeats 'We've got everything under control.'
Caption: We’ve totally got things under control
Read the full 3-panel strip
Based on the original story on theregister.com.
Editor's note
OpenAI's decision to cut off internet access during model training shows they're taking steps to prevent their AI from accessing and potentially spreading harmful or outdated information. Meanwhile, Anthropic's latest apology highlights the ongoing challenge for AI developers to keep their creations safe and reliable as the technology evolves.
Rate this comic
Read the script
Panel 1
OpenAI, known for cutting-edge technology, disables internet access during model training to avoid overexposure to harmful content, while Anthropic continues to issue apologies for their model's behavior. The irony of a company known for internet-based AI disabling the internet during crucial operations, while another company is constantly apologizing for their AI's behavior.
OpenAI: A leading AI research company known for creating advanced AI models, like ChatGPT.
Anthropic: Another AI research company working on developing ethical AI, recently in the news for addressing issues with their own models.
Panel 2
*Inside a dimly lit, futuristic lab with computer screens and a large server rack. A researcher, looking tired and hunched over, is typing away on a keyboard. The room is filled with wires and monitors showing various AI training processes.* Researcher (whispering to themselves)
Researcher: If only the internet could be trained like a pet.
Panel 3
*The same lab, but now a large poster with a QR code and a message reads: The researcher is pointing at the poster, looking bewildered and scratching their head.* Researcher (to themselves)
Narrator: DISABLED: INTERNET ACCESS DURING TRAINING.
Researcher: How do you teach an AI to be cautious without letting it explore?
Panel 4
*Outside the lab, on a sunny park bench. A different researcher, looking more relaxed but still holding a clipboard, is reading a statement. A crowd of people in casual wear are gathered, listening intently.* Researcher (into microphone) Crowd Member (smiling awkwardly) Researcher (sighing)
Researcher: We’re deeply committed to ensuring our AI learns the right way... even if it means unplugging it.
Crowd Member: So, no more internet for the smartest machine in the world?
Researcher: For now, that’s the safest way to keep it from... well, causing more issues.
Highlight: How do you teach an AI to be cautious without letting it explore?
Comments