In the short-lived news app Artifact, that was one of my favorite features. It was done on demand, and if a high portion of early viewers asked for a rewritten title, the rewrite would become the default for future viewserves.

In the Artifact implementation, the LLM was specifically prompted by the app to summarize the article with an honest, non-clickbaity title. In Google’s case, they claim they are prompting the LLM to title the link to better tempt the searcher to click on it based on what they were searching for. Kind of the opposite. Yes, LLMs could do what you say, but that doesn’t seem to be how Google is setting it up.

the unsupported, internal format - that allows for pretty much everything - is in some way exploitable

It prevents AI summaries of the video from including all the video content. Obvious solution: remove the information missing from the summary from the video itself. Now they match! /s

My commute to work includes a main city road with two lanes each way and a turning lane, and sometimes there is a school bus that stops.

The left isn’t immune: you see similar framing with genetic engineering and nuclear power generation. Rejection of nuance is systemically fertilized in widely consumed right-wing media in a way that the current left-wing media doesn’t, but the underlying psychological susceptibility is common to all humans and not particular to ideology.

Any improvement needs to happen at the system level, not by villianizing individuals as “right wing, can’t handle shades of grey”.

The current iteration of agentic AI technology used by Logitech is little more than a glorified note-taking bot capable of summarizing meetings and “generating” the occasional idea.

Given that most humans hate note-taking and avoid it, but it has a lot of value as a meeting output, getting a machine to do it makes sense.

I also heard a podcast where a consulting company couldn’t get their client contact to make any decisions because he wanted his CEO to review, but she had a busy schedule and was never available. The consultants trained an AI on this CEOs writings, and presented it to their client contact. The model was convincing enough the client felt comfortable making decisions. I thought that was interesting, and this article refers to something similar with models of stakeholders.

Not in my lifetime. I do hope leaders emerge who can harness the modern levers to our brains in a constructive way, but even if that started next year at the midterms, the decades of work ahead just to get back to the institutional knowledge and regulatory protections that existed last year exceed my life expectancy. The hope is that happens so my nieces and nephews will benefit.

I’ve occasionally been part of training hourly workers on software new to them. Having really, really detailed work instructions and walking through all the steps with themthe first time has helped me win over people who were initially really opposed to the products.

My experience with salaried workers has been they are more likely to try new software on their own, but if they don’t have much flexible time they usually choose to keep doing the established less efficient routine over investing one-time learning curve and setup time to start a new more efficient routine. Myself included - I have for many years been aware of software my employer provides that would reduce the time spent on regular tasks, but I know the learning curve and setup is in the dozens of hours, and I haven’t carved out time to do that.

So to answer the question, neither. The problem may be neither the software nor the users, but something else about the work environment.

The skills of both writing useful minutes and prioritizing actually sending them out are frustratingly rare. An average meeting with five or six people has even odds of not including someone with both of those skills. I can see where reliably having a mediocre AI summary might be an advantage over sometimes having superb human-written minutes and sometimes having nothing.

Different people and relationships can have different solutions that work for them. That’s OK!