Anthropic debuts a GPT-4 rival as AI giants say the tech’s still untrustworthy
Catchin’ a ride on the hype train… Alphabet- and Amazon-backed Anthropic said on Monday that it had released a new version of its AI chatbot. The biz said the bot (Claude 3) outperforms OpenAI’s GPT-4 and can do things like summarize book-length chunks of text and analyze images. Anthropic’s not the only company selling a shiny new bot. Microsoft’s been pushing its ChatGPT-powered Copilot for Outlook, Word, and Teams, with early testers paying a monthly fee of $30/user for access.
Nicked by the cutting edge… AI’s being sold as the next big thing to optimize worker productivity, but even the companies doing the selling have said the tech still has serious issues. Last month, both Anthropic and Google told The Wall Street Journal that their AI bots could generate false statements framed as facts (though companies have pushed their bots as tools for news research). Meanwhile, the VP of product management at Google’s AI unit said you can’t trust the model’s output. AI’s introduction to the workplace hasn’t gone smoothly:
Mistranscript: One of Copilot’s marketed abilities is summarizing Teams meetings, but users observed that it fabricated quotes from nonexistent employees (who’s “Bob”?).
Power-no-Point: One user shared on X that Copilot’s attempt to make a slide presentation was riddled with errors (verdict: “a mess”).
Release first, ask questions later… Techies are racing to unleash their latest and greatest AI offerings while cautioning customers that their output’s not reliable. The lack of polish may be one reason the tech has struggled to fully catch on. Microsoft’s splashy AI addition to Bing has so far failed to make a dent in Google’s search dominance. AI companies are moving at lightning speed to be first, but gaffes may hurt their goal of building trust: two-thirds of business execs said AI’s still in the hype phase.