Golden gate Claude was so preoccupied with being a bridge, it totally forgot about guardrails. They basically gave it ADD with hyper focus capabilities.
Thanks for the video. I think ai chat with voice or text chat is most important. For example: I witnessed a student having trouble verifying if the US educational system had a military background. The student was about to trust the information was 'fabricated' because he couldn't find academic information on the topic. I'm confident the student couldn't find his answer due to his age, (personal data on the internet) and the latest (search/biases) of his Google history. I asked him to enter Google Scholar from other countries in his research, to change the question, and to add 1970s article archives available. There, in seconds he found the information he was seeking, wasn't 'fake news' after all. Maybe we have two major ethical issues to consider as we move along 'simplifying' (?) LLM performance. Governments are destroying paper archives, and libraries are becoming social hubs. No more books to back up data. (Conspiracy theorists are questioning the amount of deleted data from platforms such as Google). Who are most of the developers/programmers setting up the root of information and its structures in LLM? Aren't they as young as the student I mentioned above? Is trusting AI at this stage a good idea? What if ethics regarding LLM may not improve soon enough to keep up with other new trends in LLM progress.
So when we think about our identities, we have all this subjective data front and centre. When we put something else front and centre in an LLM, it identifies as that. When a child of 2 years old first gets obsessed with a thing or person, they often begin identifying as that thing. My nephew chose Humpty Dumpty. What can we point at in the AI's mind, and say "tha't's you"? Will it love itself?
Surely it is not beyond engineers to create a system that is trained on high quality sources, and can fact check through a majority voting system of experts.
Golden gate Claude was so preoccupied with being a bridge, it totally forgot about guardrails. They basically gave it ADD with hyper focus capabilities.
Thanks for the video.
I think ai chat with voice or text chat is most important.
For example: I witnessed a student having trouble verifying if the US educational system had a military background. The student was about to trust the information was 'fabricated' because he couldn't find academic information on the topic. I'm confident the student couldn't find his answer due to his age, (personal data on the internet) and the latest (search/biases) of his Google history.
I asked him to enter Google Scholar from other countries in his research, to change the question, and to add 1970s article archives available.
There, in seconds he found the information he was seeking, wasn't 'fake news' after all.
Maybe we have two major ethical issues to consider as we move along 'simplifying' (?) LLM performance.
Governments are destroying paper archives, and libraries are becoming social hubs. No more books to back up data.
(Conspiracy theorists are questioning the amount of deleted data from platforms such as Google).
Who are most of the developers/programmers setting up the root of information and its structures in LLM? Aren't they as young as the student I mentioned above?
Is trusting AI at this stage a good idea? What if ethics regarding LLM may not improve soon enough to keep up with other new trends in LLM progress.
Thanks IBM, this MOE is a really good analysis of the Ai race..❤4rmZambia 🇿🇲
So when we think about our identities, we have all this subjective data front and centre. When we put something else front and centre in an LLM, it identifies as that. When a child of 2 years old first gets obsessed with a thing or person, they often begin identifying as that thing. My nephew chose Humpty Dumpty. What can we point at in the AI's mind, and say "tha't's you"? Will it love itself?
#TyroneBostickIBM
You guys need Christopher Lafayette in this discussion.
Surely it is not beyond engineers to create a system that is trained on high quality sources, and can fact check through a majority voting system of experts.