Gemini is my favorite, but it does seem to be prone to “breaking” the flow of the conversation.
Sharing “system stuff” in its responses, responding to “system stuff”, starts sharing thoughts as responses, responses as thoughts, ignoring or forgetting things that were just said (like it’s suddenly invisible), bizarre formatting, switching languages for no reason, saying it will do something (like calling a tool) instead of doing it, getting into odd loops, etc.
I’m guessing it all has something to do with the textual representation of chat state and maybe it isn’t properly tuned to follow it. So it kinda breaks the mould but not in a good way, and there’s nothing downstream trying to correct it. I find myself having to regenerate responses pretty often just because Gemini didn’t want to play assistant anymore.
It seems like the flash models don’t suffer from this as much, but the pro models definitely do. The smarter the model to more it happens.
I call it “thinking itself to death”.
It’s gotten to a point where I often prefer fast and dumb models that will give me something very quickly, and I’ll just run it a few times to filter out bad answers, instead of using the slow and smart models that will often spend 10 minutes only to eventually get stuck beyond the fourth wall.
I think that’s exactly why they’re not including timestamps. If timestamps are shown in the UI users might expect some form of “time awareness” which it doesn’t quite have. Yes you can add it to the context but I imagine that might degrade other metrics.
Another possible reason is that they want to discourage users from using the product in a certain way (one big conversation) because that’s bad for content management.
I generally agree that they are garbage at producing code beyond things that are trivial. And the fact that non-techies use them as “fact checkers” is also disturbing because they are constantly wrong.
But I have found them to be very helpful for certain things, for example I can dump a huge log file and a chunk of the codebase and ask it to trace the root cause, 80% of the time it manages to find it. Would have taken me many hours otherwise.
But as for today, have we all just collectively decided to pretend that the LLMs we have are capable of writing good software?
I use LLMs a lot in my workflow, I probably spend a whole day per week learning and fiddling with new tools, techniques, etc. and trying to integrate them in all sorts of ways. Been at it for about a year and a half, mainly because I’m intrigued.
I’m sorry but it still very much sucks.
There are things it’s pretty good at, but writing software, especially in large brownfield projects, is not one of them. Not yet, at least.
Most of them are, yeah. There are a lot of Idea Guys on here, who are in love with the idea that they no longer need effort or skills to Create Their Vision. If they can just prompt hard enough, success will come rolling in.
Did either of you read the article? You seem to be arguing against a point it doesn't make. Tools like Claude Code are entirely capable today of one-shotting tiny bespoke web apps that do a narrow set of things for an audience of one.
The article isn't talking about "large brownfield projects" or people wanting "success [to] come rolling in". It's about people making little apps for themselves, for personal enjoyment, not profit.
I’m naturally pretty pale and don’t get much sunlight, I feel like I look like shit unless I get just a little bit of tan. What most people would consider just a healthy looking “baseline”. It also puts me in a better mood although that may be entirely psychological.
When I was younger I used to intentionally tan for short durations, but now I realize that’s harmful so I just embrace the cave gollum look
I am white as paper, probably one of the palest people and I live in Asia and often get comment that I have the dream skin. While back at home my parents were teasing me about being a ghost and doctors asking am I sick. Interesting how it changes on cultural basis
I think it’s more than just cultural. Yes, it’s definitely a factor, and there are cultures and there were times where paper white was considered beautiful.
But I think on some level we naturally associate severe paleness with being sick or non-social.
Not sure really I am not an expert on this, where I live now and look at some of the wealthy people, they are extremely white like on purpose. Some of the leading politicians too. In fact, it's a bit difficult to find a very dark skinned celebrity or a powerful politician here, there are some but not many at all.
To me personally, I like naturally tan skin (like Asian natural skin) > natural white skin > artificial tanned skin > heavy tanning. Tanned white people just do not look good to me.
If you asked someone else where I live now, I bet answer would be different
To me, something like RFK Junior skin looks disgusting. I always wince when I see a picture of him, like you could make that into leather bag.
The mood is probably part light and part vitamin D. The latter can be supplemented. The former can be reproduced with a full spectrum bright lamp or brief sun exposure in the morning.
I mean sort of but you should probably just get some sun if you can. There’s such a thing as too much tanning, sure, but getting no sun is not healthy either.
Exposing large amounts of skin to the sun has other health risks when it is freezing outside. :)
Vitamin D deficiency is very common in Canada particularly during winter. The government recommends that everyone intentionally seek out vitamin D rich foods, or to take a supplement.
I think the underlying assumption is that we are “real”, meaning our existence is grounded in some undisputed “reality”. So if what we perceive as the universe isn’t real, then there has to be some other real universe that is simulating it in some way.
Sharing “system stuff” in its responses, responding to “system stuff”, starts sharing thoughts as responses, responses as thoughts, ignoring or forgetting things that were just said (like it’s suddenly invisible), bizarre formatting, switching languages for no reason, saying it will do something (like calling a tool) instead of doing it, getting into odd loops, etc.
I’m guessing it all has something to do with the textual representation of chat state and maybe it isn’t properly tuned to follow it. So it kinda breaks the mould but not in a good way, and there’s nothing downstream trying to correct it. I find myself having to regenerate responses pretty often just because Gemini didn’t want to play assistant anymore.
It seems like the flash models don’t suffer from this as much, but the pro models definitely do. The smarter the model to more it happens.
I call it “thinking itself to death”.
It’s gotten to a point where I often prefer fast and dumb models that will give me something very quickly, and I’ll just run it a few times to filter out bad answers, instead of using the slow and smart models that will often spend 10 minutes only to eventually get stuck beyond the fourth wall.
reply