{"id":15329,"date":"2023-06-02T22:25:53","date_gmt":"2023-06-02T22:25:53","guid":{"rendered":"https:\/\/www.insightpartners.com\/?p=15329"},"modified":"2023-06-02T22:25:53","modified_gmt":"2023-06-02T22:25:53","slug":"ai-news-june-2","status":"publish","type":"post","link":"https:\/\/www.insightpartners.com\/ideas\/ai-news-june-2\/","title":{"rendered":"AI-list: Need somebody with the human touch"},"content":{"rendered":"

AI is a rapidly developing sector that Insight has invested more than\u00a0$4B since 2014<\/a>. We\u2019re currently tracking 23,315 AI\/ML startups and ScaleUps. Naturally, our team has seen a lot, and the discussions around this topic generate a lot of opinions.<\/em><\/p>\n

Here\u2019s a look behind the curtain into some of what we\u2019re reading, sharing, and discussing across the team at Insight lately. Think I missed something?\u00a0Email me<\/a> and tell me all about it. See the archive of all our AI roundups here<\/a>.<\/em><\/p>\n

\n

Less is more?<\/h2>\n

Last week, I declared less is not more<\/a>. This week, the New York Times is on it<\/a> with “BabyLMs.” The challenge? Train a language model on only ~100M words, or approximately the number of words a 13-year-old encounters. This might make the AI sound more human, the theory goes.<\/p>\n

\n

Making bigger and more capable AI chatbots requires processing power that few companies possess, and there is growing concern that a small group, including Google, Meta, OpenAI, and Microsoft, will exercise near-total control over the technology. Smaller LMs can be more accessible, proliferate innovation, and specialized for niche uses. They can also present challenges for regulation.<\/p>\n

To be continued.<\/p>\n<\/div>\n

\n

AI Is Writing Code Now. For Companies, That Is Good and Bad.<\/a><\/h2>\n<\/div>\n

As GenAI enables a lot more workers to code, CIOs worry about managing code complexity and tech debt. Via WSJ (and Insight’s own network), we’re hearing a lot of this sentiment, along with real questions on costs and governance. Seems like a problem that could be solved…with AI?<\/p>\n

A quote from our Insight AI chat proffers this: “You could imagine people creating tests and then doing AI code generation based on human-created tests, and then using the tests to check the quality of the code gen. Like an AI supercharged version of Test Driven Development (TDD).”<\/p>\n

An apple a day keeps the AI away<\/h2>\n

Academia (along with various disciplines of “knowledge work,” specifically marketing and software development) has arguably been one of the most disrupted areas as ChatGPT gained widespread adoption. Educators have been frazzled trying to address student use of AI tools and what it means for the future of assessment.<\/p>\n

From the WSJ <\/a>(paywall, sorry, but his speaking tour<\/a> is free, and his 2018 TEDxCMU<\/a> talk about math is worthwhile), educator and Carnegie Mellon mathematician Po-Shen Loh is evangelizing “knowing how to solve problems, and which problems to solve,” and leaning into the humanity of it all. “The value of ingenuity will only get stronger.”<\/p>\n

Meanwhile, the Twitter academic hivemind has been experimenting with how to beat ChatGPT at its own game.<\/p>\n

\n

So I followed @GaryMarcus<\/a>‘s suggestion and had my undergrad class use ChatGPT for a critical assignment. I had them all generate an essay using a prompt I gave them, and then their job was to “grade” it–look for hallucinated info and critique its analysis. *All 63* essays had<\/p>\n

\u2014 C.W. Howell (@cwhowell123) May 27, 2023<\/a><\/p><\/blockquote>\n