I’ll be doing a signing & discussion for my book Co-Intelligence at the Harvard Coop Bookstore this Thursday at 6pm. If you are interested, here is the link: https://
eventbrite.com/e/author-event
-ethan-mollick-tickets-886462762987
…
RESEARCH
-
Ethan Mollick Co-Intelligence Book Signing at Harvard Coop
By
–
-
LMSYS Performance Reflects RLHF Choices, Not General Capabilities
By
–
perhaps bc lmsys performance is less a judgment on general capabilities than it is of rlhf choices i dug into the data a little for curiosity sake – always worth remembering how short (and single turn) most of these queries are.
-
New Book on AI Engineering with Foundation Models
By
–
I’m excited to share that I’m working on a new book about building applications with foundation models! AI Engineering builds upon Machine Learning Systems Design, but with a focus on large scale, ready made models. The book covers: – The new AI stack (e.g. how it differs from
-
Gemini 1.5 Launch: Google’s Latest AI Model Arrives
By
–
Confirmed, that was Gemini 1.0 – Gemini 1.5 just showed up right near the top!
-
Open Weights 70B Model Reaches GPT-4 Performance Level
By
–
Surprising because this is showing an open weights 70B model at GPT-4 level (for any prompt I may wish to ask)
-
Understanding Tokenizers: The Vocabulary of AI Models
By
–
I wrote this explainer on tokenizers last year: https://
simonwillison.net/2023/Jun/8/gpt
-tokenizers/
… Effectively they're the vocabulary used by the model -
AI Models Remain Vulnerable to Adversarial Attacks
By
–
Sadly not: that paper concludes with "Finally, our current models are likely still vulnerable to powerful adversarial attacks." More of my notes here:
-
LLM Safety: Instruction Hierarchy Defense and Team Hiring
By
–
LLMs process text from multiple sources and may face conflicting instructions. We teach our models to follow instructions from the highest priority input, giving better defense against attacks. Our Safety Systems team is hiring:
-
Instruction Hierarchy Advances LLM Robustness Against Prompt Injections
By
–
Introducing the Instruction Hierarchy, our latest safety research to advance robustness for prompt injections and other ways of tricking LLMs into executing unsafe actions. More details:
-
Meta’s 15T Token Training Strategy Preference
By
–
would prefer they just trained on 15T tokens like meta