try-working

we will soon figure out how to update the engrams with new information

"Software is solved. Bugs are not yet solved"

I think I'll use the DeepSeek Search API instead and "google it" that way instead

DeepSeek.

this would be nice with an eSim and an agent application so I can control DSH on my M5U at home. would that be possible?

that's a good point, one search just to get to try it out. I guess I could add that.

What error did you see when you tried to log in? I'm not sure if it's displayed clearly enough, but the callback will always fail in the pop up and you'll have to copy the callback URL into the text field in the first modal, and let the worker acquire the credential.

same. bought the annual plan, second largest one, in around February to lock in the price. well, what did they do, cut quota by 80% when K2.7 launched and then even further when K3 and K3.1 launched.

Sales tax in China is attributed to the place of production rather than the place of sale, which is why provinces want to produce as much as possible

you actually have no idea what the cost is

DeepSeek doesn't really care about models, they're aiming to become a platform: https://try.works/a-follow-up-on-china's-open-ai-race

you dont need to build a plugin to use local models with DSH. you can configure models and providers freely by default.

yes you can modify ui and everything else. i have a few dsh- plugins on my github (in profile), but i havent found any external ones yet

you don't understand the article. users in china access anthropic models via relay stations. the stations scam the users by routing some requests to cheaper chinese providers instead. thats why anthropic would see mixed requests.

it doesn't matter if they're an early stage company or not. also they were a fraud company.

Completely fake. Those companies used relay stations, or third party gateways, to access Anthropic models. https://x.com/trydotworks/status/2098300730805284961?s=20

After using it for 24h in DSH, this is a very good model and much better than the previous version. Cancelling my Codex sub when 4.1 Pro is released, or maybe before then.

DeepSeek Harness allows the user/agent to build its own plug-ins and run them without restart, mostly. Pi and OpenCode were then influenced to do the same in their respective new versions. I don't know which has the better implementation.

Yes, and therefore RSI isn't RSI. You need to set up multiple environments to generate data sets, multiple RL envs and just train the models on different datasets, architectures and then compare the final results against each other. The RSI part is just generating these pipelines end to end automatically.

I have used over 40B tokens and spent over $800 on DeepSeek API over the past 30 days, mostly on V4.1 Flash.

It's good, and you can do most work with this. For complex software implementation you need to split your runs into various phases, build in verification, and use subagents so that work gets another audit and repair pass from the lead agent. You can do pretty much everything then. Frontier models can do without compelx workflows, that's the difference.

I built an STT plugin for DeepSeek Harness that uses this Whistle model as well as a larger one from Desert Ant Labs: https://github.com/try-works/dsh-stt

if you want to get a sense of what China is like, you can look at how China talks about itself: https://chinaonchina.com/

news from Chinese media, translated into English.

there is nothing to add because TFA is "not even wrong".

There are insane speed improvements for local inference going around on X right now. They've popped up the last month and week.

Tensorfold is getting 100%+ speed increases on both prefill and decode for models like Qwen 27B. oMLX has followed them and have had similar improvements in the past week.

There's lots of different techniques like letting CPU help with prefill, DFlash specualtive decoding etc.

I'm really excited for this as I'll be receiving an M5U in about a month. Expect to be running Qwen 4 27B or Flash (it's a 96gb machine), and they may come close in performance to DS 4/4.1 Flash, and should be able to hit 100 tps. Local is really becoming viable, especially considering that GPT 6.1 has been running at 20ish tps the past week.

Very strange post. I've spent 40% of my life in China. I think these comparisons are forced by someone who wants to make comparisons and find clear differences.

For social media, Xiaohongshu is about the same as international social media, except you don't have that very weirdest type of brainrot content.

in Effect, please

It's just the takeover of the continent by an ossified bureaucratic class. Their opinions and actions do not represent the European people's but quite the opposite of it.

httpd://recursive-mode.dev

I reiterate how unusual, chaotic, and fraud-filled this election has been:

1. More than 5,000 Swedish citizens have made official appeals against the legitimacy of the election 2. More than 100,000 Swedish citizens have signed a petition for a re-election due to election fraud, interference, tampering etc: https://ettfolk.se/

3. On the same site above, you can read about 500 examples from Swedish voters of fraud, influencing, interference etc

4. Accused Somali candidates have held meetings with Somali officials in Sweden and in Somalia, discussing how the Somali population should act in the Swedish election to further Somali interests

5. In the 2022 election, a muslim party received about 20-30% of votes in the muslim areas. In this election, the same party received almost exactly 0.0% votes - instead, the incriminated leftist parties increased by the same amount and even more. All voters shifted to the same 1-2 parties in an organized manner.

None of this has ever happened before in Swedish history