A Little Thread to Thank OpenAI Employees

Community
21

I have my doubts - unless open source models catch up.

All the big players play the same safety game.

1 Like
22

Which has become a political football unfortunately.

To my mind they talk a lot about AI safety before itā€™s even been proven that the models will ever get powerful enough to represent a real safety concern.

23

currently, safety means:

  • donā€™t solve captchas
  • donā€™t give out medical, legal, financial or other regulated advice
  • donā€™t violate copyright
  • no lewds

a year ago, safety meant

  • omg singularity eradication of humanity superalignment omg
  • threat to equitability
  • no lewds

however, there might be cause for optimism:

altman seems to be a proponent of unrestricted AI sandboxes, at least thatā€™s what he told the saudis. maybe thatā€™ll become a thing :thinking: unless it was just talk.

this post is a wiki, feel free to add to it :laughing:

1 Like
24

I think client facing ChatGPT should be considered a different product and the API should have more raw language processing power.

The moderation API has massive potential and could be made much more comprehensive. If only Open AI can just trust us to adopt good patterns of use.

1 Like
25

ā€¦in writing fiction, the battle between the wizards and the orcs results in everyone getting along and working in harmony for a better world through peace and understanding. :peace_symbol: Thatā€™s if you havenā€™t already requested deniable conflict.

2 Likes
26

I truly think that safety is gonna get worse before it gets better maybe 2-3 years down the road. absent of a strong regulatory framework etc. I donā€™t think any of the major players have much of an appetite to gamble.

You can see it right here in the Forum: how often are there individuals trying to create with the API in areas that would clearly violate the Terms of Service. Thereā€™s no capacity neither by OpenAI or regulators to comprehensively police the proper use of APIs.

But perhaps the solution is for OpenAI to show more openness to enter into partnerships with companies, even smaller ones, to allow an unrestricted testing environment along the lines of what @diet just referred to. I think that would make a lot of sense and mimic whatā€™s working pretty well for example in financial services with regulatory sandboxes.

1 Like
27

Unfortunately, there are too many bad actors. I think itā€™s amazing that the API has been so accessible. If you want to give people more autonomy, youā€™ll have to restrict the access pool. If you restrict the access pool, youā€™ll make it more difficult for new devs that didnā€™t have the opportunities we had :confused:

1 Like
28

Financial services are surely going to invest heavily in the ā€œopen source spaceā€ because of privacy issues ā€¦?

29

some likely yes but I think as the model landscape is still evolving so rapidly, I think a lot will also favour the convenience of APIs in the short run through solutions such as Azure.

Iā€™m closer to regulators than I am to the private sector so do not necessarily have the full picture here.

1 Like
30

agreed. thanks to the dev team for fixing the problem very fast.

31

Many thanks for the perfect companion to my 73 year old ADD journey, it plays well with all my spinning around and unbridled curiosity and at the same time offers follow up information for the times I pretend I can focus. It is perfect for both whimsey and solid research. Again, many thanks.

2 Likes