- The Daily Update
- Posts
- ChatGPT gets lazy during the holidays
ChatGPT gets lazy during the holidays
News: Democratic campaign deploys AI
Jack Shields
December 13, 2023
Welcome back!
Is it possible for AI models to experience “holiday fatigue” like humans? New research says yes — AI has “learned” to do less work over the holidays.
Today we’ll take a look at that research and more breaking news about AI-powered political campaigns and the technology’s impact on the environment. Let’s dive in.
In today’s Daily Update:
🗞️ Democratic campaign deploys the world’s first AI robocaller
🤖 ChatGPT reportedly gets lazier during the holidays
📸 Microsoft aims to power AI ambitions with nuclear energy
🚨 AI Roundup: Four quick hits
Read time: 2.5 minutes
TOP STORY
🗞️ Democratic campaign deploys the world’s first AI robocaller
Source: WTAE-TV
Democrat Shamaine Daniels hopes the world’s first AI campaign volunteer will help her win a seat in Congress currently held by Republican Scott Perry.
What you should know:
Ashley is an AI robocaller that does not use pre-recorded responses. She is capable of having an infinite number of customized one-on-one conversations simultaneously.
Her creators say she is the first political campaign caller powered by generative AI, and intend to mainly work with Democratic campaigns and candidates.
Ashley has already called thousands of Pennsylvania voters on behalf of Daniels.
The AI analyzes voters’ profiles and tailors conversations around their key issues. Unlike human volunteers, Ashley has a perfect recollection of all of Daniels’ positions and does not feel demoralized by voters who hang up.
Why it matters: Ashley is ushering in a new era of political campaigning by conducting high-quality conversations at a large scale. This comes with concerns that AI hallucinations could present voters with misinformation, and creates a legal gray area that will need to be addressed in the coming years.
RESEARCH INSIGHT
🤖 ChatGPT reportedly gets lazier during the holidays
Generated by Adobe Firefly
A new study posted on X by Rob Lynch shows that ChatGPT performs worse during the holiday season.
The details:
Last week OpenAI confirmed that ChatGPT is getting “lazier,” but did not offer an explanation for the trend.
Lynch conducted his research with two system prompts: one that told GPT-4 it was May and another that said it was December.
He found that GPT-4 produces shorter completions when it “thinks” it’s December than when it thinks it’s May.
Wharton Professor Ethan Mollick reposted the study and cited the AI Winter Break Hypothesis as a potential explanation, suggesting that AIs might perform worse in December because they “learned” to do less work over the holidays.
The relevance: Lynch’s research reiterates that AI models can inherit undesirable traits and biases from the humans who train them. This is crucial for developers to keep in mind as they build AI applications for the military and government.
BUSINESS SPOTLIGHT
📸 Microsoft aims to power AI ambitions with nuclear energy
Source: Bjoern Schwarz
Microsoft is turning to nuclear power to fuel its ventures into AI and supercomputing.
Key points:
The tech industry has a growing need for power as a single new data center requires as much electricity as hundreds of thousands of homes.
Nuclear power is carbon-free and provides electricity around the clock, but subjects developers to an expensive U.S. nuclear regulatory process.
Microsoft recently posted a job opening for a nuclear expert to help it integrate small modular reactors (SMRs) into its electricity planning.
The U.S. is among a group of countries that pledged to triple global nuclear power generation by 2050.
The significance: AI’s impact on the environment has been fairly underreported compared to other risks like AI-driven misinformation and deepfakes. Microsoft’s push to go nuclear offers a strong alternative to current industry practices.
MORE TRENDING NEWS
🚨 AI Roundup: Four quick hits
Generated by Adobe Firefly
Trend Setter: New York Times hires first newsroom leader focused on AI.
AI Biographer: Google explores AI that “knows everything about your life.”
W.A.L.T.: Stanford researchers introduce AI model that generates photorealistic and consistent video from text prompts or still images.
Cutting Ties: Sports Illustrated publisher fires CEO after AI controversy.
THAT’S ALL FOR TODAY