I don't think there's another way. Everybody is being spammed with Gen AI outputs and the only even to it seems like a paradox way to fight it is with LLMs of some kind or go around it like what the EU is trying to do with watermarking and then detect AI and have different processing for it. I kinda think we could end up there, but we are still not there.
no its not, one is you dont resolve anything for the country and therefore dont operate in it, the other is you operate in the country and the block you implemented does not work for my customer.
similar how a lot of the US pages blocked europeans from accessing because they did not want to follow the cookie laws.
no, its different, in german case they served germany as a whole but for queries that where supposed to be blocked in germany they did not serve those to germans.
so you operated in germany and your "tricky" way to follow the regulations did not work. if you dont resolve to german customers at all, like you dont exist then nobody can theoretically say you are operating in german. so that is the difference for me.
this is properly bad as you can run glm on 6000pro, at (in my eyes) pretty good quant, and a q2 on a 5090. im suprised at how good 5.3 flash is at some parts. i still think ds4.1 flash is better but it is more demanding. i hope meta pushes something in a range of 300-400b with a 15-20b active parameters. i know its a crowded space. but it would be fast and probably mostly good enough.
i run opus on medium, sonet on high, and haiku on high for implementation based how complicated is
for planing and specing and review opus is on high usually, except if if the task is contained.
did not try to run it on low, will have to try it.
I have both, 6.1 sol is no where near 5x more efficient then opus in day2day. Without numbers i would not even entertain that though and with numbers i would look closly if its the same code. I put it to work on the same repo so i have somewhat of a comparison. Not to mention that opus 5.5 is way better in most of the tasks. Currently openai gives you less for worse idk how did they think that could work.
you are not false, but there is still difference. its just that the better models are correct more of the time and will better validate its own steps. glm sometimes will understand the plan start implementing and then forget part of it and then say it finished. or then take a wrong turn somewhere and not correct. but they will all happily proclaim they are correct till you question it.
I would suggest you to hook fable or 5.6 to check it regularly and its work because it gets lost easily on stuff it was not trained on. I'm doing some custom inference engine optimization and it's a workhorse but it can easily lose its way and if you don't recheck it you will get wrong answers in the end.
I also have a legacy pro plan and the only limitation is if you are trying to work in the morning from Europe because you are in the 3x usage overlapping China time but after 12 or so you basically can run it at least for me at least 3 parallel sessions all the time.
it was not the first model served for free, i remember grok and others beeing free on openrouter but they never had this popularity because they where not good enough.
To be fair, there is no 3 turns that I don't have to jump in into what Opus 5 is doing. There is either some regression or my prompting skills are so much worse now. Flash is not perfect and honestly some things depend on how big context do you keep. So I'm keeping like a really short context with my flash, but it works okay, even though it has a tendency to overthink, and yeah, I run it always in max effort mode.
I will give it a try, but from the benchmarks it never exceeds the DS4 flash benchmarks by significant margin and And I feel that the throughput that you will get on those machines or what I'm getting with my local hosted flash will be so much worse that it's not worth it.
i run flash v4 at 2bit, its pretty great and on my tests against full model It didn't lose any capabilities. It just was thinking more. So you don't have the same efficiency.
You don't care what were you trying to build but for a quick me alone Linux app I used Flutter.
Honestly it's my go-to when I want to have a quick native app on any platform but dont want to bundle electron.
It's not that they don't care, it's just that the protocol behind it was never designed for it.
Compared for example to email where when you want to send email to somebody, you discover its MX settings on the DNS level and then send it directly and the receiver can verify are you allowed to send those emails. With phone calls it's basically a network of interconnects so I'm sending the call directly to my provider but that provider to reach the destination has multiple steps and they have to trust each other with the numbers signalling. Compare it to a regular snail mail where everybody can write anything as a sender.
What most of the providers are switching to now is for enduser the case is if the phone number that you want to signal is not registered with us, you are not allowed to signal it. Which is fine and covers most of the cases but imagine that you are a business and have multiple providers And you have boughta number block with one provider and another one is offering you better minute price for outgoing calls. Or you want to have a forwarding to your cell phone number. Or you want to signal your number when you're calling prospects from CRM. The regulation usuall differs from country to country, but most allow you to signal any number that you own with any provider that you use. And this is correct, but the problem is verification.
They all are pretty similar. And if it was copying anything, I would say more inspiration from claude desktop then codex. The Zcode is more like codex.
> People familiar with OpenCode internals (if you are on the OpenCode dev team I assume this doesn’t include you) might have objected to my python3 example above.
Cache Misses are pretty bad, i have a locally running deepseek v4 flash, i have tuned it now to have 1100-1300 prefill. Its not great but properly useable. Imagine having a session with already 100k and half of it has to be prefilled it would be waiting minutes with worse numbers. And if you are paying by the token, for hosted models, you are wasting money.
Diagnosed with ADHD, ultimately does not change anything for me even through i had the same idea as you. Reason is that i can now start even more stuff in parallel. And some part of them get finished more before i can just prompt more when in focus, but instead of finishing i add more features.
same, but you need more then 100k of hw to run something like kimi k2.6 for a bigger team. on the other hand there is a ds4 flash that you can run on a macbook with 128gb ram. an that one is perfectly usable for a lot of tasks.
The problem is not website, the problem is discovery and discovery is on Instagram, TikTok, and social networks. You don't have any incentive to build a website for a regular audience. What you might do is build an audience on a social network and then try to move them to a website.
But at that point you're big enough to build it properly.
depends, a super small one finetuned to do function calling instead sending it to big model and waiting, instead, you ask for a revenue in last month, i do a small llm function call -> show results. some bigger ones, analysis, summary, classification.
what is great with smaller ones, and im looking at 2b, 4b is you can get a huge throughput with just vllm and a couple of consumer gpus.
what i usually do is basically distillation of a big one onto smaller one.