They made a kind of strange decision with Qwen3.8 27B, the template defaults the reasoning_effort to xhigh. I found if you set it to medium it doesn’t just sit there churning forever.
162 karma · joined March 20, 2023
Looking at your examples - ChatGPT, you need some way to get an otherworldly sized dataset before you can train a model, and the fact that you also happen to have to write a web interface for the chat looks like a footnote in comparison - Amazon Prime, have to create a distribution empire - Crude Oil Tracking, have to get raw data from somewhere, I don’t know the space well but I’d be shocked if they didn’t have some moat around the data source, or they even have a hand in collecting it - Netflix, you can solve all the hard problems of video streaming and still have nothing people want to watch - Bloomberg Terminal, this one is maybe the closest to being replicable with just good software? I’m sure building the data sourcing for it would still be the “hard part”
If you broaden your scope from SaaS to “software that makes money” the most obvious are social networks, but then your problem becomes how to actually monetize it, since you more than likely can’t charge the users for it.
IIRC it does just set a time limit on thinking
The truth is probably more that the various money sinks in our society are starting to add up, things like healthcare, legal protection, licensure, compliance, rent (business or personal), even just having appropriate work vehicles, fueling them, compensating people for the time spent sitting in traffic to come across town to your house. Somehow you’re paying for all of that when someone’s livelihood is installing your mini split. A lot of those costs have grown faster than wages, if you try to point to a reason why it’s different today than 20 years ago. More people looking to make a quick buck without doing any work or providing any real value, and more people succeeding.
It does feel like it must be the engagement version of junk food, though.
“The algorithm” knows perfectly well what you want to watch. So it seems inevitable to me that eventually “the algorithm” just generates an endless stream of content fine tuned to you.
Keep the IP while you run the servers and sell the game, when you are no longer interested in running the servers, drop a tarball of code. It doesn’t need to be simple, it just needs to be possible and complete. The community can take care of it. If the IP is valuable, then you have your incentive to keep the game playable until it isn’t valuable to you.
I fundamentally believe the same is true of SWEs, you can’t get more out of people than they have long-term by telling them they ought to be “sprinting,” yet if you ask me that’s the concept being pushed.
Even in the case of something like a credit card company where it’s not technically a monopoly, you risk them all coming to the same conclusion.
Essential for-profit service combined with at-will service is simply a recipe for failure.
This is fundamentally an incentive problem. Whether MegaCorp reaches this decision through an AI or a team of people who made a mistake, interpreted a rule differently than intended, missed context, whatever, they have been allowed to set up the system in a way which has all the incentives in their favor. They don’t need you and further, helping you costs them money. Yet we allow them to be the only game in town and still say it’s at their convenience to give you service.
The focus in the modern world needs to be on re-incentivizing companies to do the right things, AI or otherwise.
I’ve personally watched the enshittification of too many proprietary tools to ever build something I care about on top of one today, especially something which becomes so fundamental to the design of your application like a GUI toolkit.
I know it sounds crazy, like you’d never bother forking your GUI framework anyway even when it’s open source. But I worked on an application built in C++ Builder, at a company with enough engineering talent and willpower that we would’ve forked the compiler internally to fix its problems and limitations had we been granted access to the source. Instead, I got to watch the product get held back by it year after year, hoping that Borland would care.
It’s a good trick to have in your back pocket if you’re given an unprivileged user on a compute node and want to make use of modern tools.
Plus mistral-nemo in particular has a large context window, so you can cook up some shell scripts to throw a bunch of context into the buffer before your question. One I use a lot takes the name of a manpage and a question about it, then the LLM has the whole manpage to reference.
In that case I see no problem with the main sponsor company not wanting just anyone to come and make money off their work. They might accept contributions if offered up, but they are not hoping to gain much from them.
That’s in stark contrast to a project like Linux where it much more heavily relies on outside people getting into the development cycle.
To your point actually, I think it can be sticky for an open source maintainer of a small project when someone comes along and tries to be a more active contributor and treat your project like the latter when you’re really intending it to be more like the former. There’s no great signal of what type of open source you’re intending to create apart from saying “I don’t really want significant contributions” in your readme.