🥟 Chao-Down #335 Amazon plans AI overhaul of Alexa with a monthly subscription, Google scrambles to remove weird AI results from search, TikTok prepares US-version of their recommendation algorithm
Plus, a look at how China is building its own versions of AI chatbots.
After a much-needed vacation with family, I’m back! What did I miss in the world of AI? Apparently quite a lot, especially one interesting story around AI models and their favorite numbers.
Engineers at data science firm Gramener found that popular large language models like OpenAI’s GPT-3.5 Turbo and Anthropic’s Claude 3 Haiku show human-like biases when asked to pick random numbers between 0 and 100.
Apparently, GPT 3.5 Turbo really likes the number 47, Claude 3 Haiku loves 42, and Gemini is infatuated with the digits 72. Even when the engineers increased the “temperature” (the knob that controls variability in LLM responses), the results were the same.
Just like people, the AI models tended to avoid extreme high or low numbers or double digits (no 22s, 55s or 66s). And that makes sense given the fact that these models are trained on mountains of human-authored texts.
So what does this mean? Does AI really know about what is random? Certainly not.
AI responses are imitations of human behavior, devoid of consciousness or understanding, selecting the most probable token to answer a given query. But the fact that they are able to regurgitate clusters of commonly occurring numbers speaks more to the fact that we have our own biases towards randomness.
Maybe we don’t like things to be too random after all.
-Alex, your resident Chaos Coordinator.
What happened in AI? 📰
Amazon plans to give Alexa an AI overhaul, monthly subscription price (CNBC)
Google scrambles to manually remove weird AI answers in search (The Verge)
TikTok preparing a US copy of the app’s core algorithm, sources say (Reuters)
China’s ChatGPT: why China is building its own AI chatbots (Nature)
Chat Xi PT? China’s Chatbot Makes Sure It’s a Good Comrade (WSJ)
Anthropic's position on AI safety is helping it lure talent away from OpenAI (Fortune)
Always be Learnin’ 📕 📖
2024 Gen Z and Millennial Survey (Deloitte Global)
How Generative AI Is Remaking UI/UX Design (a16z.com)
AI Agents & New Age of Software (Substack)
Projects to Keep an Eye On 🛠
lencx/ChatGPT: 🔮 ChatGPT Desktop Application (Mac, Windows and Linux) (Github)
glanceapp/glance: A self-hosted dashboard that puts all your feeds in one place (Github)
apache/superset: Apache Superset is a Data Visualization and Data Exploration Platform (Github)
The Latest in AI Research 💡
LoRA Learns Less and Forgets Less (arxiv)
Observational Scaling Laws and the Predictability of Language Model Performance (arxiv)
Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking (arxiv)
The World Outside of AI 🌎
How Ozempic Is Changing Diabetes Treatment (The New York Times)
A Lot of What You’ve Heard About Cortisol Is Wrong (WSJ)
The Mall Food Court Near You Is Making a Comeback (Bloomberg)
1 in 7 Gen Z credit card users are ‘maxed out’ (CNN Business)
Why tech billionaires are trying to create a new California city (CBS News)
Why L.A.'s tech scene has fallen short of the Silicon Beach hype (Los Angeles Times)
One Last Bite 😋
