SAFETY - AI Dynamics

AI Supply Chains: Liability and Downstream Inference Consequences

By

–

14 July 2023 19h21

But what are the consequences of these supply chains on downstream inferences? How do we determine liability when things go wrong? Read our early discussions On AI Deployment: http://
gradientscience.org/supply-chains W/
@andrew_ilyas @cen_sarah @aleks_madry @LVidegaray @imstruckman

→ View original post on X — @mit_csail

14 July 2023

Generating Benchmarks for Factuality Evaluation of Language Models

By

@_akhaliq

–

14 July 2023 3h22

Generating Benchmarks for Factuality Evaluation of Language Models paper page: https://
huggingface.co/papers/2307.06
908
… Before deploying a language model (LM) within a given domain, it is important to measure its tendency to generate factually incorrect information in that domain. Existing

→ View original post on X — @_akhaliq

14 July 2023

Anthropic Launches Researcher Access Program for Claude 2 API

By

@anthropicai

–

14 July 2023 0h06

We are excited to support research into safety and societal impacts via our Researcher Access Program. We aim to provide subsidized API access to our frontier models, including Claude 2, to as many researchers and academics as we can.

→ View original post on X — @anthropicai

14 July 2023

Mode Collapse in Large Language Models: Book Output Example

By

@mit_csail

–

13 July 2023 23h27

Here’s actual example of mode collapse in an LLM, where “book” is repeatedly outputted.

→ View original post on X — @mit_csail

13 July 2023

10 reasons to worry about generative AI

By

@tamaramccleary

–

13 July 2023 21h10

10 reasons to worry about generative #AI https://
bit.ly/3lC3L2j

→ View original post on X — @tamaramccleary

13 July 2023

Model Performance Improvements Across Demographics Without Fairness Tradeoffs

By

@aiatmeta

–

13 July 2023 18h17

Early experimental results show that our approach improves model performance on all demographic groups in our evaluation datasets, with the largest gains in respect to accent inclusivity — demonstrating that improved fairness doesn’t mean a performance tradeoff. 2/3

→ View original post on X — @aiatmeta

13 July 2023

Privacy-Preserving Approach Improves Speech Recognition Fairness

By

@aiatmeta

–

13 July 2023 18h15

Today we're announcing a new privacy-preserving approach to improve fairness & robustness of automatic speech recognition systems. This unique approach lets researchers improve ASR performance without relying on demographic data. More info

→ View original post on X — @aiatmeta

13 July 2023

Cruise Driverless Service Reliability Issues in San Francisco

By

@rodneyabrooks

–

13 July 2023 6h50

Driverless not a deployed SF "service". Watched @cruise creep towards me for 13 minutes tonight. As it went right past me app informed me of service difficulty, so should try again later. 60 seconds later I was in an Uber. @cruise service success down to 1 in 3 over last 3 weeks.

→ View original post on X — @rodneyabrooks

13 July 2023

Sarcastic critique of AI alignment solution timeline claims

By

@rasbt

–

13 July 2023 6h18

Congrats! You should join forces with the other company who just announced that they will solve AI alignment in <4 years. So you can both work on “solving the universe” once your are done with these more near-term projects.

→ View original post on X — @rasbt

13 July 2023