Creative and important work from @JessicaRumbelow and @SoC_trilogy at SERI MATS. Their research is ongoing — see the Part 2 update here: https://
lesswrong.com/posts/Ya9LzwEb
faAMY8ABo/solidgoldmagikarp-ii-technical-details-and-more-recent
…
RESEARCH
-
Creative work from Rumbelow and SoC_trilogy at SERI MATS Part 2 update
By
–
-
GPT-3/ChatGPT Anomalous Tokens Elicit Bizarre, Evasive, Hallucinatory Generations
By
–
"SolidGoldMagikarp": Prompting GPT-3 / ChatGPT to repeat any of several hundred anomalous tokens elicits bizarre generations — described by researchers as variously "evasive," "hallucinatory," "insulting," "ominously humorous," and "religiously themed." https://
lesswrong.com/posts/aPeJE8bS
o6rAFoLqg/solidgoldmagikarp-plus-prompt-generation
… -
Open Source Repository for API Research Documentation
By
–
i have often thought that all this stuff should be compiled in an open source repo somewhere since there must be a bunch of people independently doing this same api research. not sure its possible to cover this level of detail tho
-
Google Brain PyGlove Library Shares ML Ideas as Code
By
–
Google Brain Extends Its PyGlove Library to Easily and Scalably Share ML Ideas as Code https://
syncedreview.com/2023/02/08/goo
gle-brain-extends-its-pyglove-library-to-easily-and-scalably-share-ml-ideas-as-code/
… -
Facebook’s Major Open Source Contributions: PyTorch and React
By
–
Imagine being Facebook and building both pytorch and React and then…
-
Commonly heard term: retrieval augmentation or just retrieval
By
–
I usually hear it called retrieval augmentation or just retrieval
-
Andrej Karpathy Rejoins OpenAI to Build AI Future
By
–
Some personal news: I am joining OpenAI (again :)). Like many others both in/out of AI, I am very inspired by the impact of their work and I have personally benefited greatly from it. The future potential is especially exciting; it is a great pleasure to jump back in and build!
-
ChatGPT Bing Search Accuracy Issues with Personal Information
By
–
Bing search with ChatGPT knows that I’m a reporter, and is only in error about my age, my birth date, my current employment status, and what I look like (not sure who this photo is of.) // @MSFTResearch @OpenAI // #ChatGPT #AI #machinelearning #OpenAI
-
Microsoft Research AI Systems Plagued with Errors ChatGPT
By
–
True, it will be riddled with errors. // $MSFT @MSFTResearch @OpenAI // #chatgpt #AI #OpenAI
-
LLM Fact-Checking and RLHF: Critical Stack Components
By
–
The recent news around Google Bard's incorrect fact in the ad points to 2 increasingly critical parts of the nascent LLM stack: – LLM Expert Verification—experts in the loop constantly fact-checking LLMs
– RLHF—encourage the model to bullshit less, and rely more on exact facts