SAFETY - AI Dynamics

Constitutional AI: LLMs as Reward Model Annotators

By

–

02 February 2023 20h56

Worth checking out constitutional AI from @AnthropicAI where LLMs can be the reward model annotators by prompting well. The choice then comes to paying for GPU or Phillipines.

→ View original post on X — @aravsrinivas

2 February 2023

ChatGPT Passes US Medical Licensing Exam Without Medical Training

By

@therundownai

–

02 February 2023 17h46

Recently, ChatGPT passed the United States Medical Licensing Examination (USMLE). This test normally takes 4 years of medical school and 2+ years of clinical rotations to pass. http://
medrxiv.org/content/10.110
1/2022.12.19.22283643v2.full
…

→ View original post on X — @therundownai

2 February 2023

GPT-4 Writer Style Mimicry Detection Challenges

By

@willknight

–

02 February 2023 16h03

I've wondered about that myself. Surely a big part of it is simply that you can instantly see elements of other artists style in imagery, and it's much harder to sense that copying and remixing in text? If gpt4 can very effectively mimic particular writers it might change.

→ View original post on X — @willknight

2 February 2023

Attention Is All You Need Impact on AI Progress and Safety

By

@clementdelangue

–

02 February 2023 15h32

It does count for the progress & safety of the field! If Google didn't release "Attention is all you need", AI would be decades away from where we are!

→ View original post on X — @clementdelangue

2 February 2023

Mercedes-Benz Obtains Nevada Certification for Drive Pilot Level 3

By

@actuiafr

–

02 February 2023 11h01

Drive Pilot, @MercedesBenz's SAE level 3 system, certified in Nevada https://actuia.com/actualite/drive-pilot-le-systeme-sae-de-niveau-3-de-mercedes-benz-certifie-au-nevada/
… #AI #artificialintelligence #car

→ View original post on X — @actuiafr

2 February 2023

Would AI prefer many small paperclips or one large paperclip?

By

@sama

–

02 February 2023 2h46

would an AI realistically want a lot of little paperclips, or one big paperclip? (reason for question to follow later)

→ View original post on X — @sama

2 February 2023

Safer E2E Conversational AI Through Value Sensitive Design

By

@cohere

–

02 February 2023 2h03

(11/12) Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design
Authors: A. Stevie Bergman, @gavin_does_nlp
, Shannon Spruit, @dirk_hovy, @em_dinan
, Y-Lan Boureau, Verena Rieser

→ View original post on X — @cohere

2 February 2023

A Watermark for Large Language Models

By

@cohere

–

02 February 2023 2h02

(4/12) A Watermark for Large Language Models
Authors: @jwkirchenbauer
, @jonasgeiping
, Yuxin Wen, Jonathan Katz, Ian Miers, Tom Goldstein

→ View original post on X — @cohere

2 February 2023

Neutral AI defaults and user preference alignment strategies

By

@sama

–

02 February 2023 0h32

we are working to improve the default settings to be more neutral, and also to empower users to get our systems to behave in accordance with their individual preferences within broad bounds. this is harder than it sounds and will take us some time to get right.

→ View original post on X — @sama

2 February 2023