It was another bananas week in the world of AI.
AutoGPT was released some weeks ago, but this week, marketing use cases began to roll in for the autonomous AI service. It “will literally perform the role of a marketing professional with minimal supervision” Gene Marks wrote in Forbes.
From building apps, social media community management, optimising customer data to market research, AutoGPT began to deliver outcomes which far exceed the expectations one might have for a platform only weeks old.
Unlike the chat-based AI platforms, AutoGPT does not require a prompt or human supervision: instead, users provide it with a goal or task, and the AI figures out the steps required to complete the objective.
This requires problem-solving, memory and internet connectivity. Critics warned against “continuous” or god mode, in which the AI continues to work without human supervision or involvement.
Truth is in the AI of the beholder
Having called for a six-month pause on generative AI development just days ago, Elon Musk this week claimed he was building a new AI called TruthGPT, an endeavour the clown-shoe wearing billionaire said would be a “maximum truth-seeking AI that tries to understand the nature of the universe.”
Content without consent
As OG internet forum Reddit announced it would charge for the use of its content to train future AIs, and developer platform Stack Overflow followed suit, Washington Post journalists, along with researchers from the Allen Institute for AI analysed the online content used to train Google’s Large Language Model. They found a plethora of offensive, biased and problematic source material had been used, and copyrighted material appeared more than 200m times in the Google data set. ChatGPT has not shared information on its training sources.
Ghosted by the machine
Social media platform Snapchat expanded “My AI” to its entire 750 million subscriber base. The company initially made it available only to paying subscribers. Commentators believe the intention is to optimise the available data for a resurgence of Snap’s advertising platform, but it also raised concerns about the role AI will play in further disconnecting a pathologically online generation from real-world interactions.
Is My AI taking McGinn’s side?
Brain Deep Mountain High
The two rival Google AI teams of DeepMind and Brain were brought together as a single unit called Google DeepMind. The move signalled a more aggressive approach to the search giant’s role in the AI wars.
The Financial Times revealed Google’s plans to use generative AI to create advertisements based on material produced by (human) marketers. The new AI, which also powers the Bard chatbot, will enable Google to produce more sophisticated campaigns and repurpose brand assets at scale - another significant disruption for agencies. Meta, which owns Facebook and Instagram, has been using an AI-based platform to this effect for some time.
mAIcrosoft
The paid links are not clearly delineated.
Newton’s law
On a closing note for the Week in AI, in this week’s Hard Fork podcast, host Casey Newton revealed the impact of ChatGPT on his peers. When tired of hearing someone talk, the buzz phrase de jour is to tell them to “Stop generating”.
Don’t forget - early bird tickets to humAIn are on sale now.
12 July 2023
NSW Teachers Federation Auditorium, Surry Hills
Early Bird Expires 17 May 2023