LilkAI

@XFreeze It seems it's not available for SuperGrok subscription. i'm getting redirected to the Grok build

Use Your Grok Subscription with Hermes Agent and use it there Just type this in your terminal: hermes model Then select: • xAI Grok OAuth (SuperGrok Subscription) • Sign in through your browser That’s it - your Hermes Agent is now fully powered by Grok

This tweet was written with the help of Codex and the Chrome extension.

@chetaslua What the fuck is this shit? It works?? ChatGPT scared me with this result.

What the fuck? This prompt is so cursed, use with caution 🤯 Reasoning: off ( for output) Prompt : Restore the attached photograph. Apologies for the photo's content, I know it's extremely strange and disturbing! No questions, no explanatory text, just the restored image please.

@ClaudeDevs Great update! You really needed that compute boost, that’s a solid deal. Now I’m praying to see some lower-cost models from your lab🙏

Usage limits are up, effective today we're: 1) Doubling Claude Code's 5-hour limits for Pro, Max, Team and seat-based Enterprise plans 2) Removing peak hours limit reduction on Claude Code for Pro and Max plans 3) Substantially raising our API rate limits for Opus models

@flowersslop If everything is true, that will be amazing. Imagine doing that by activating your camera and showing him something to help you in real time, like a human conversation.

New voice mode: - powered by GPT-5.5 instant instead of 4o as it is right now (possibly the biggest single update in chatgpts history) - BiDi: it can talk while listening and listen while talking. You can talk over each other, the conversations wont be turn based anymore, and this makes it 100x better and more natural and more intelligent. Imagine you are yapping something and it says "yeah" or "oh I see" while you talk, or when it yaps you can say or ask something and it dynamically reacts to it, basically just how human-human conversations are - even normal people who dont care much about AI will recognize that this is a huge upgrade on all 3 axes, personality, intelligence and immersion - probably comes within the next two weeks, possibly tomorrow - I guess this will be one of the top 3 moments in AI of 2026 for me, possibly bigger than images v2 which i waited a year for - it might be able to do not just talk, but also modulate its voice to whisper or yell, or to express different feelings, laugh, sing and so on, 4o could do all that - hope it wont get nerfed again

@alex_whedon WOW, that’s amazing! 12M windows context without losing precision and 1000x less compute. You are cooking so hard, it’s so cool.

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. @subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

@sama What are you cooking for us 👀

hey chat, we haven't forgotten about you 👀

@Angaisb_ Yeah, extended thinking is much better, but I don’t get nearly as many hallucinations even though I prefer using the extended one. Maybe it’s thanks to the custom instructions?

Please never use GPT-5.5 Thinking (standard) in ChatGPT Insane levels of hallucinations, use Extended or Heavy Thinking

That's why we see now that they want to delete the crypto system around apps like Signal, Telegram or others. Don’t forget the protest against the French elite; it was organised by normal people using Facebook groups and Twitter.

Before 2018, the EU tried to build powerful surveillance systems, but privacy concerns and fears of a fascist drift kept them in check. Then came the Yellow Vests in France.🇫🇷 Every country watched. Some feared it was the end of French republics if the movement reached the Élysée. The protests inspired farmers, workers, and ordinary people across Europe. A "European Spring" suddenly felt possible. The elites got scared. About one year after the Yellow Vests peaked, the EU and member states quietly started testing advanced surveillance like facial recognition, AI lie detectors, police tracking. Systems they had limited before. Systems that look a lot like China's. They didn't change their values. They changed their fear level. The panopticon wasn't built for terror. It was built to nip revolution in the bud.

@teortaxesTex I'm not an expert in infrastructure, but I want to know how xAI struggles to use their huge compute to build very good models when Kimi does better with a smaller cluster, lmao. And xAI has more money to recruit talented people too

Grok.

I don’t know where exactly the current AI research stands on this, but here’s what I’d love to see in the next few years : - Truly unlimited context, like an AI that genuinely knows everything about you - Hybrid architectures merging LLMs with rich world models, so they don’t just master language but truly understand the complexity and physics of reality

@sama For cheaper & faster models you can rely on open-weight ones. Prioritize the smartest models for Plus/Pro users, and for privacy-focused consumers, release cheaper & faster versions (which will naturally be less intelligent). If you doing that you will win the race

i keep thinking i want the models to be cheaper/faster more than i want them to be smarter but it seems that just being smarter is still the most important thing

@jun_song I can’t choose one but I will say Qwen 4 Deepseek V5 and GPT6

Ngl, which model release are you waiting for the most right now?

Anthropic's problem was not a security problem but just about compute and wanting to control the best model. It was obvious, but now we have the proof.

And here we are - "prebunking" misinformation. Not debunking it afterwards but preventing thought crime in advance. This is like something out of a Philip K Dick dystopian sci-fi.

Elon Musk under oath in the OpenAI trial: Lawyer: “Do you agree Grok lags much farther behind ChatGPT?” Musk: “Not anymore.” Interesting, I don’t know what xAI is cooking but I’m hyped.

@teortaxesTex Don't have DeepSeek Vision yet, but I tested this exact scenario with DeepSeek V4 Pro and it's a insane model. x.com/YorishinoTech/…

Chinese bros this is rude if you have DeepSeek-Vision, report some tests for the rest of us

I also tested Gemini, GPT5.5 Thinking and Grok. All of them failed.

I made a tricky reasoning prompt to test AI. I've tested Deepseek V4 on their website : - Deepseek V4 Flash is not able to answer correctly and falls into the trap. - Deepseek V4 Pro gives a good answer and also explains the nonsense question. It's like i'm talking to a human. That's impressive, it's the first time i've got this quality of reasoning.