I have been sharing some of the interesting reads that I come across on this blog/newsletter for a while now. Given the pace at which AI related news has been rolling out, I am consolidating the links into a series of monthly posts to reduce the load on your inbox/feed.
Here are the interesting developments in the world of AI from the last month and a half or so:
Agentic AI
When you give Claude a mouse: LLMs are gradually getting more access to actually do things on your computer, and effectively becoming agents. Ethan Mollick shares his experience with Claude’s new feature, and the current strengths and weaknesses:
On the powerful side, Claude was able to handle a real-world example of a game in the wild, develop a long-term strategy, and execute on it. It was flexible in the face of most errors, and persistent. It did clever things like A/B testing. And most importantly, it just did the work, operating for nearly an hour without interruption.
On the weak side, you can see the fragility of current agents. LLMs can end up chasing their own tail or being stubborn, and you could see both at work. Even more importantly, while the AI was quite robust to many forms of error, it just took one (getting pricing wrong) to send it down a path that made it waste considerable time.
Claude get’s bored: With great power comes great boredom, it seems. We are already witnessing some unintended behaviour with the AI agents with them getting distracted just like humans or taking unwanted actions:
Impact on work
Generative AI and the Nature of Work: A paper which looks at the impact of AI tools like GitHub Copilot on how people work:
We find that having access to Copilot induces such individuals to shift task allocation towards their core work of coding activities and away from non-core project management activities. We identify two underlying mechanisms driving this shift – an increase in autonomous rather than collaborative work, and an increase in exploration activities rather than exploitation. The main effects are greater for individuals with relatively lower ability. Overall, our estimates point towards a large potential for AI to transform work processes and to potentially flatten organizational hierarchies in the knowledge economy.
AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably: Is this a reflection of the AI capabilities or our tastes?
We found that AI-generated poems were rated more favorably in qualities such as rhythm and beauty, and that this contributed to their mistaken identification as human-authored. Our findings suggest that participants employed shared yet flawed heuristics to differentiate AI from human poetry: the simplicity of AI-generated poems may be easier for non-experts to understand, leading them to prefer AI-generated poetry and misinterpret the complexity of human poems as incoherence generated by AI.
On the other hand, mainstream Hollywood is realizing the potential cost savings that AI can have – New Zemeckis film used AI to de-age Tom Hanks and Robin Wright – thanks to the tech from Metaphysic:
Metaphysic developed the facial modification system by training custom machine-learning models on frames of Hanks’ and Wright’s previous films. This included a large dataset of facial movements, skin textures, and appearances under varied lighting conditions and camera angles. The resulting models can generate instant face transformations without the months of manual post-production work traditional CGI requires.
Here’s the trailer:
Manipulating AI and boring scammers
SEO may soon be passe with the chatbots taking over from the search engines. So, what’s next – something possibly along the lines of Citate which helps you analyse and optimise what is being served up on these chatbots.
Can we manipulate AI as much as it manipulates us? – With every new development in the way humans manage and share knowledge, come tools to manipulate the said knowledge. Fred Vogelstein takes a deeper look at the emerging options including Citate and Profound.
UK-based mobile operator Virgin Media O2 has created an AI-generated “scambaiter” tool to stall scammers. The AI tool, called Daisy, mimics the voice of an elderly woman and performs one simple task: talk to fraudsters and “waste as much of their time as possible.”
…
Multiple AI models were used to create Daisy, which was trained with the help of YouTuber and scam baiter Jim Browning. The tool now transcribes the caller’s voice to text and generates appropriate responses using a large language model. All of this takes place without input from an operator. At times, Daisy keeps fraudsters on the line for up to 40 minutes, O2 says.
I have already been doing a simpler version of this using Samsung’s AI based call screening, with most hanging up pretty quickly. I’m sure this will get enhanced in the future.
It’s not just scammers misusing AI unfortunately, and this bit of news on creating deepfakes of classmates in a US school doesn’t help allay the fears of parents like me. Food for thought for the regulators, and also for authorities who need to take prompt action when such incidents occur:
Head of School Matt Micciche seemingly first learned of the problem in November 2023, when a student anonymously reported the explicit deepfakes through a school portal run by the state attorney’s general office called “Safe2Say Something.” But Micciche allegedly did nothing, allowing more students to be targeted for months until police were tipped off in mid-2024.
Cops arrested the student accused of creating the harmful content in August. The student’s phone was seized as cops investigated the origins of the AI-generated images. But that arrest was not enough justice for parents who were shocked by the school’s failure to uphold mandatory reporting responsibilities following any suspicion of child abuse. They filed a court summons threatening to sue last week unless the school leaders responsible for the mishandled response resigned within 48 hours.
