SAFETY - AI Dynamics

Aligning Large Language Models with Human Values

By

–

06 September 2023 22h58

Aligning Large Language Models with Human: A Survey by Yufei Wang, Wanjun Zhong, Liangyou Li, Fei Mi, Xingshan Zeng, Wenyong Huang, Lifeng Shang, Xin Jiang, Qun Liu

→ View original post on X — @cohere

6 September 2023

Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback

By

@cohere

–

06 September 2023 22h58

Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback, by @StephenLCasper @alxndrdavies @causalclaudia @sociotiose @jeremy_scheurer @javirandor @FreedmanRach @tomekkorbak @davlindner @5kovt @CRSegerie @MicahCarroll
, et al.

→ View original post on X — @cohere

6 September 2023

Digital Data Guardrails: First Step in AI Regulation

By

@terence_mills

–

06 September 2023 20h00

Digital data guardrails are the first step in regulating AI
#AI #AIio #BigData #ML #NLU #Futureofwork @TopCyberNews @SpirosMargaris @MarshaCollier @MHiesboeck
@MHcommunicate @Fisher85M @MikeQuindazzi @NealSchaffer
http://
ow.ly/Y1Ge30sx0Te

→ View original post on X — @terence_mills

6 September 2023

AI and Risk Management Discussion with EFIPM

By

@yuhelenyu

–

06 September 2023 17h34

Such a pleasure chatting up with @efipm on #AI and #RiskManagement. https://
youtube.com/watch?v=ZNvc2J
4bPwM
… @NACD #BoardDirector #CEOs

→ View original post on X — @yuhelenyu

6 September 2023

Google Contractor Pay Disputes Impact Bard Quality Control

By

@jordannovet

–

06 September 2023 16h28

this sounds great- a Google subcontractor at Appen who works on Bard is not given enough time to check facts in the chatbot's answers, while Appen, which is in the middle of a turnaround, isn't paying the $15 an hour that Google requires for contractors https://
cnbc.com/2023/09/06/app
en-which-helps-amazon-and-google-train-ai-is-reeling.html
…

→ View original post on X — @jordannovet

6 September 2023

Chinese AI Model Spreads COVID-19 Misinformation About US Origins

By

@mfordfuture

–

06 September 2023 15h26

Chinese ChatGPT competitor "reckons that covid-19 originated among American vape-users in July 2019; later that year the virus was spread to the Chinese city of Wuhan, via American lobsters."

→ View original post on X — @mfordfuture

6 September 2023

AI-Powered Phishing Threats Set to Escalate Dramatically

By

@steve_rosenbush

–

06 September 2023 14h58

Eyal Benishti, Ironscale CEO: “Just imagine business email compromise and targeted phishing at the same volume as we experience spam, because that’s what will happen.” #AI #cybersecurity #generativeai #phishing

→ View original post on X — @steve_rosenbush

6 September 2023

FACET: New Fairness Benchmark Dataset for Vision Models

By

@aiatmeta

–

05 September 2023 21h31

Last week we released FACET, a new comprehensive benchmark dataset for evaluating the fairness of models across a number of different vision tasks, constructed of 32K images from SA-1B, labeled by expert annotators.

Read the paper ➡️ https://t.co/OoYV2eiSYX pic.twitter.com/9Q7uk59TvT
— AI at Meta (@AIatMeta) 5 septembre 2023

Last week we released FACET, a new comprehensive benchmark dataset for evaluating the fairness of models across a number of different vision tasks, constructed of 32K images from SA-1B, labeled by expert annotators. Read the paper https://
bit.ly/3EotZvg

→ View original post on X — @aiatmeta

5 September 2023

AI Conversational Abilities Don’t Guarantee Safe Decision-Making

By

@stanfordhai

–

05 September 2023 17h03

AI has demonstrated unprecedented capabilities such as engaging in natural conversations, but this doesn't mean we should trust these models to handle sensitive decision-making tasks. "It’s just not there yet,” says HAI faculty affiliate Sanmi Koyejo.

→ View original post on X — @stanfordhai

5 September 2023