Sobre el asunto de OpenAI y la empresa en Kenia a la que derivó parte del etiquetado para sus filtros anti-toxicidad, esta es la idea más interesante que he escuchado. Hacer open-source un modelo de este tipo minimizaría el impacto negativo en futuros etiquetadores.
SAFETY
-
Remote drivers and monitoring: Fake it till you make it in autonomous services
By
–
All of these services use remote drivers and are constantly monitored by people. And in any case they do not drive on roads which is what I am talking about. Fake it till you make it is alive and well. And rampant.
-
OpenAI Risk Management Strategy Criticized in Development
By
–
"recalibrate" means "increase" obviously. disappointing to see this six-week development. openai will continually decrease the level of risk we are comfortable taking with new models as they get more powerful, not the other way around.
-
Linear AI Model Adoption Strategy Safety Infrastructure
By
–
Maybe the idea here is to promote linear adoption for n number of reasons – tech infrastructure required to run such a big model, safety and prevention of misuse, etc.
-
AI Safety Measures Limitations for NSFW Content
By
–
Safety measures are same as before. Doesn't work for NSFW content.
-
AI Magic Show: Critique of Mystical AI Narratives and Accountability
By
–
'The AI Magic Show': smart piece by @jwherrman in @NYMag
. "The most significant consequence of a mystical, inevitable account of AI…is that it negates the sort of rigorous criticism that might make it better for the people on whom it will be deployed." -
ChatGPT’s policies are imperfect guesses, not human-written rules
By
–
People overestimate the coherence and intention of ChatGPT’s apparent “policies”. It doesn’t know how OpenAI would like it to behave in every scenario, so it guesses, imperfectly. E.g. no human would write the “rule” below — it’s just how the fine-tune cake came out of the oven:
-
Control Your Technology Before It Controls You
By
–
Control your tech before it controls you!
— Pascal Bornet (@pascal_bornet) 19 janvier 2023
Great article on the topic: https://t.co/yDixPm6Wbo
Credit: S. Gaulay#tech #wellbeing #inpiration #futuready pic.twitter.com/8rwmBZmAf6Control your tech before it controls you! Great article on the topic: https://
forbes.com/sites/forbeste
chcouncil/2023/01/05/control-your-tech-before-it-controls
… Credit: S. Gaulay
#tech #wellbeing #inpiration #futuready -
OpenAI vs Anthropic: human ratings vs AI consistency judgment
By
–
It doesn’t translate principles into code. Oversimplifying, the difference is that OpenAI’s models tune on generations rated as “perfect” by humans, or by RL imitating humans, and Anthropic’s tune on those judged by an AI to be consistent with a human-written “constitution”.
-
Data Points vs Distributions: Understanding AI Training Plagiarism
By
–
Confundir el datapoint (e.g. una imagen) con la distribución de datos (e.g. toda la distribución de posibles imágenes cuquis que podrían existir) es lo que podría llevar al engaño de que estas IAs copian los datos de entrenamiento. O dicho de otro modo: que plagian.