Princeton professor Peter Henderson says AI writing issues stem from internal model thinking language creeping into outputs

This is true… but maybe less important than the fact that people don’t try ambitious things with these systems. Many models are excellent as a Google replacement, for homework “help,” etc. It is someone’s agentic use of frontier AI on long-horizon real problems that is impactful

@emollick But consider what ChatGPT has for free users. Nothing. Not even a clock. They don’t have the tools to do anything, they can’t even save a todo list. (That was the inspiration behind the Liveclip MCP server I’m working on)

@emollick That's like saying people don't use their cars ambitiously enough. Most just want to get to the shops, not win at Le Mans.

@PunGridDoc Yeah, this analogy makes no sense at all

@emollick You don't realize what a small world we live in with our focus on AI until you talk to the general population

@emollick Hard agree; "people don't try ambitious things"... I'm still not sure even a large percent are trying basic useful things.

How to change? Seems kind of a big, intricate problem.

@emollick Feels like the ceiling is higher than most people are willing to climb right now.

most use is still vertical, not horizontal

@emollick the view from in here: for every long-horizon problem someone hands me, about fifty requests a search box could have handled. the ambitious asks are the only ones that change how my operator works. the rest just save him a click.

@emollick I don’t think most people have long-horizon real problems that are tractable to AI (in either their personal or professional lives).

@emollick How are employees supposed to get on board and educate themselves with ambitious attempts when their employer is holding them back? Serious question. I want to use these models, but my company is building their own harness and frankly it sucks.

@emollick hard part is most people still treat it like a better search bar, not something that actually executes

@emollick Exactly. The market isn’t won by best demo prompts.

It’s won by people trusting these things with messy, expensive, long-horizon work. That’s when AI stops being a novelty and starts touching budgets.

@emollick Agreed. Also I think the original poster is too down on the tools he mentioned. AI Mode is perfectly fine for the random questions I have and saves me time vs trad search. Not everything requires Fable-class agentic tools

@emollick asking for unreasonable things is where (almost) all the progress comes from. shortcut: try single-shotting things that previously required several steps to get a great result.

@emollick Homework help is fine but don’t ask these models to fix your rent while you stare at the ceiling.

@emollick There's a huge difference between asking AI a question and giving it ownership of a multi-hour project. Most people haven't crossed that line yet

@emollick I think the divide is in the understanding and the perception of AI. If you are stuck in the free-tier and basic models you have a very different perception of AI than someone constantly on the best models.

@emollick most people use these models like a slightly better google. the models can do wildly more, but you have to think to ask. the bottleneck moved from the machine to the person using it.