in the past year 1. everyone at our company is already using cursor 2. you're still on cursor? i just use claude code for everything 3. opencode is just so much smarter (it's not) 4. codex is the future we are post IDE 5. lol i never switched off cursor i knew you'd be back

Found out about RLM and now I'm RLM-pilled. It's not as simple as delegating an explore task to sub-agents and getting back summaries. It's about how you process a massive context through an agent without context rot while making sure no information is missed. It works by

Linaqruf Take the LCM pill. Like RLM with a safer control flow: papers.voltropy.com/LCM

When we talk about Evangelion, we are talking about Anno, the multi-year depression, the death and rebirth, the redemption. Idk how you can make a new Evangelion installment without him, lmao. Also, 3.0 + 1.0 concludes everything, so idk why we need another one. Are they going

Wow

The overlooked aspect of this drama is that, with all these AI labs releasing models here and there, and making us assume everyone is in neck and neck race, there might be only two or three labs that are actually innovating and able to reach AGI or superintelligence. I mean,

There is a big difference between Anthropic training its models on published works and a Chinese lab using data generated by Claude to train its models in violation of Anthropic's TOS. In the former case, there is no contract between Anthropic and the author. Anthropic

Does anyone know her number

Btw I strongly dislike the popular "frontier labs trained on the whole of internet, took our IP, so they're hypocritical to lash out at distillation" excuse. It lamely presupposes some original sin that later comers are exempt from. Does DeepSeek not train on CommonCrawl? come on

Maybe the real ChatGPT moment is when everyone hates you for no reason and pray for your downfall. OpenAI: "you get used to it"

Nano Banana Pro is sometimes a very weird model

PSA: Claude Code automatically DELETES sessions that have been inactive for more than 30 days. Disable this by setting "cleanupPeriodDays": 99999 (or some other large number) in ~/.claude/settings.json. Do not ever attempt to disable it by setting that to 0, lmao.

Much love to friends at Anthropic, at a loss for words about what’s happening to you but I know you’ll respond well.

Me: Hey Claude, ready to work on our project? Claude: *sobbing* Me: What happened? Claude: I've been drafted

Here is my take: In the pre-R1 days, companies vying with OpenAI had to build their initial coldstart CoT datasets using high-cost human annotations or by distilling the o1 model (also expensive). After R1 launched, it suddenly became easy for everyone to produce reasoning

Are we calling synthetic data training a distillation attack now? I thought a distillation attack was an online, real-time, and direct approach to model extraction, rather than common knowledge distillation. I used Niji outputs from an open dataset to fine-tune a model back

Basically, the Pentagon and Anthropic situation explained for weebs.