Listening to Lex Fridman’s latest OpenAI conversation with Sam Altman was interesting. Below are some of the points we all should be aware of and maybe dive in deeper on.
Here are some key points, from the video linked in the comments below:
Ai is on a continual curve. We won’t know when, and there won’t be a moment to point to and say, “there” that’s when it all changed.
Per the conversation Sam Altman and Lex Fridman had. ChatGPT-4 models seem to be able to learn ‘something’ and what that something is, is hard to define, however, what Sam and the OepnAI team are seeing is a form of reasoning. The thought is that by ingesting human knowledge, the model at a certain point is gaining what could seem to be wisdom. You can test this out, and see a similar effect when using multiple prompts within the same chat. This, in turn, seems to show reasoning which is remarkable!
I personally liked the point that Sam made regarding the why of ChatGPT. He stated the goal is to enable humans by aligning to the value, and utility and driving ease of use.
There was a vague but interesting call out by Sam, where he said that the science of human guidance… and the ability to predict the outputs are more scientific than could have been imagined.
The good and how we as humans benefit from these models.
ChatGPT-4, is bringing Nuance back to the world. This was agreed on and I feel the same. To Lex’s point, ‘it’s like a breath of fresh air”
The OpenAI team is seeing that Better alignment = Better capabilities & vice versa.
In the end, I will leave you with this:
Within the conversation, there was, what I would call, a call to action or a call for help. When talking about alignment. Sam stated,
“We have yet to discover a way to align a super powerful system”. Note that the current RLHF – (Reinforcement Learning Human Feedback) does have a defined process, the issue seems future-focused.
Further in the conversation, Sam also stated…
There is no one set of human values or no one set of values for human civilization.
What’s going to have to happen is we will need to agree as a society on very broad bounds of what these systems can do! Maybe we tailor them a bit like GDPR – (General Data Protection Regulation) does in different countries or regions.
I for one vote we do just that and start a council that serves not only the above function, but others as well to provide better alignment, clarity, and transparency to the many up and coming with and around AI, AGI, and beyond.
If you read this far then who’s up for it, or agrees it is needed?
#ai #chatgpt4 #openai #aiethics
Notice: The views expressed in this post are my own. The views within any of my posts, or articles are not those of my employer or the employers of any contributing experts. this post? Click icon on the top right of my profile to be notified when I post.
