AI Top-of-Mind Nov 7
Top-of-mind are some cool enhancements announced for ChatGPT which will allow custom ‘mini’ GPTs, as well as their creation of an app store to support this. If the iPhone and Android were app 1.0, maybe this is 2.0?
Also new are enhancements for image re-creation via more sophisticated prompting. As Wei Mao describes, one uploads an image to ChatGPT, it creates a prompt, and then DALL-E3 will re-create the image to the best of its ability. Results are amazing.
Not to bore you with too much OpenAI news, GTP-4 has added checks to avoid some forms of hallucinations based on illogical prompts. Reminds me of web exploits based on malformed code.
Still on the hallucination front, the NY Times yesterday looked at the tendency of Chatbots to create their own version of reality. From the article:
OpenAI’s technologies had the lowest rate, around 3 percent. Systems from Meta, which owns Facebook and Instagram, hovered around 5 percent. The Claude 2 system offered by Anthropic, an OpenAI rival also based in San Francisco, topped 8 percent. A Google system, Palm chat, had the highest rate at 27 percent.
Maybe we need an updated version of the Turing Test, a PARRY-like interaction, the reports of Google’s LaMDA notwithstanding.
And for those with full access to ‘The Information,’ a well-curated AI startup database. And with all the buzz around Microsoft, we don’t want to leave out AWS from the Generative AI party. Topics include Anthropic and Bedrock.
Turning to social, yet another call for better control, this from the EFF, given the ease of spreading misinformation, sometimes generated via AI. From the article:
In moments of political tension and social conflict, people have turned to social media to share information, speak truth to power, and report uncensored information from their communities. There are steps that social media platforms can take to increase the likelihood that their sites are places where reliable information is available — particularly during moments of conflict.
