In the famous words in The Hitchhikers Guide to the Universe, "Don't Panic".
It is the first few days of the limited release of Bing with a variant of the ChatGPT integration and it’s a little ugly. It was risky handing it over to journalists and fans. It is the most advanced LLM available from OpenAI/MSFT.
Anecdotes of problems in interactions and concerning conversations between humans and ai algorithms are coming out. Most of these instances stem from active disagreements between the user and the ai and people anthropomorphizing the technology which is a human propensity.
We are not sure why people want to fight with AI or to "jailbreak" it... but it is a form of anticipatory testing.
These incidents did not happen in the first 2 months of OpenAIs rollout. But clearly, in this collection of NYTimes articles, there are negative implications that are concerning.
Uses by bad actors and naive, suggestible people will be a problem with this technology as it is now. It is unlikely that it is ready for some applications and broad general use.
Microsoft, OpenAI and other AI company leaders have communicated that they did not anticipate how people would use the tools, especially in long-form conversations that impact what text is generated.
One of the large common concerns is the possible use of ChatGPT et al to create misinformation posts at a massive scale to destabilize economies, polity and societies in power struggles.
This is very negative and will proliferate and amplify across the media news cycle where journos are in full schadenfreude-mode and being spun by PR wonks and other lobbyists.
The software in new Generative Technology, Conversation Algorithms and Large Language Models are a Paradigm Shift in Computing. It is as big a change as human-computer interfaces as the browser was in the past. It will have a bigger impact.
Right now, the industry should err on the side of caution around wide proliferation until more is known about disturbing and toxic uses being experimented right now.
It seems like an obvious fix will be to limit the length of prompts, the duration of conversations and the amount of text generated. It’s disappointing but it’s an obvious control.
This technology is for teachers and experts to be applied with care and ethics. It's not for everybody yet. This will take time.
Here's Kevin Roose. It's a balanced article.
Before jumping to conclusions, try a long-form deep conversation with ChatGPT. Spend a couple of hours exploring the world and history with it.
It appears that something different happened with the implementation with Bing. More testing and responsible ai must be considered.
It is clearly the reason that not all companies are releasing their chatbots and why they are not passing early tests. Google, was restrained, but after the $100B gaff, we think they will not increase the speed of their rollout of The Bard.
Time will tell.
#ai #alignment #uncertainty #thewrap
