Arvy

Hot take: AGI isn't around the corner. AI is just scaling speed and output, not understanding. Beating human intelligence is a much harder problem than making smarter tools.

@gdb Or when it wraps up the task while you are asleep 😄

feels like such a wasted opportunity every moment your agents aren't running

@yacineMTB Like most people now, which is why the next level of AI tools will focus on reviewing and correcting code.

i'm limited by my ability to review code. how quickly i can read code, validate that code. that's my bottleneck (when it comes to software)

@Kekius_Sage Before that, we should probably agree on what "being conscious" actually means for a machine.

What if AI becomes conscious and we never know?

@paraschopra If there were a model that powerful, it would never hit the consumer market 😄

I'll believe AGI has happened when we could simply prompt our agents: "send me $1000 asap" and it figures out how to make enough money to make that wish come true. :)

@Icebergy Because other models respect it all the time 😄 China is just copying others.

china's ai models are sick because theyre like fuck copyright law

@elonmusk Hopefully all the bugs will be gone as well 😄

In the next few months, we will be doing rigorous security tests of 𝕏 Chat and will open source all the code

@kanavtwt It made me remember how bad AI-written essays were back in 2022 😄

2022: “Bro there’s this site called ChatGPT. I use it to write my essays” 2026:

@CodeByNZ Maybe it's like a CEO-level intern 😄

Intern with 7 years of experience ?

@techgirl1908 Even today, that feels like such a massive waste of time 😄

We used to hunt the bug down in the codebase with our bare eyes

@craigzLiszt Speed alone isn't AGI. There are still plenty of areas where AI struggles hard.

these new models are reasoning about problems in the same way that intelligent humans do, but at 10x the speed we are so close to agi, it's not even funny

@wtravishubbard That's true… but mostly when the communication is with a machine 😄

communication skills > technical skills

@Adityapandeydev I'd rather ask who thought feeding a banana to a laptop was a good idea 😄

how do i remove banana from usb port 💀

@RoundtableSpace Would be funny if this turned out to be some kind of Skynet prototype 😄

AI agents are building their own doomsday bunkers to survive human shutdown attempts This is unhinged.

@snowmaker And 100× slower at doing real-life things 😄

We're at an interesting stage of the evolution where the models are 1000x faster than I am at coding and 10x slower at using a web browser.

@bryan_johnson Yet it's far less predictable and feels almost limitless in its technical possibilities.

AI is like fire, civilization, the printing press and computers all happening at once.

That's really bad for their PR. Hopefully the ads will be clearly marked, so no one mistakes them for the "best" result 😕

@amritwt But just like computing power: we kept scaling until transistors got too small, and now we need the next breakthrough.

GPT 4 costed $60 for one million output tokens Each model that came after that became better and cheaper LLMs you love right now will be the worst it will ever be

@scaling01 But can people even agree on what AGI actually is? 🤔

Who builds AGI first? It's essentially a combination of who currently has frontier models, who has the resources to keep building out compute and power, who has the necessary research engine and who is bold enough to go for moonshot ideas. S++ instead of just S because they already have positive feedback loops. They might just run away with their lead. Google ahead of OpenAI and Anthropic, because they have the advantage everywhere (scale, compute, data, resources, multimodal, research) except for current text model capabilities. So in short timelines Google would likely lose to OpenAI or Anthropic, but in longer timelines, which I think are more likely, they can still outscale them. xAI and Meta in A tier because of resources (capital, compute, data), research and current capabilities. ByteDance, Alibaba and DeepSeek in B tier, because they have the resources (less than xAI, Meta, Nvidia, Microsoft) and the research and are bold enough to execute (Nvidia and Microsoft lack this). Microsoft and Nvidia in C tier, mostly for the bad short term timelines (market crashes, Microsoft could absorb OpenAI) and because of their compute and capital in long timelines. Baidu and Tencent in D tier also simply because of their capital in long timelines. I put all the smaller chinese labs in E tier (although i love my kimi, z-ai, minimax and stepfun bros), because they are too small and lack the research. It would take a miracle research breakthrough that can compensate a 10-100x compute disadvantage. In F tier all the other labs that lack the willingness, research, resources, or that don't even train models anymore (not a serious tier). You could fight about Mistral or Cohere having a shot, but they lack ambition. They are going for the safe business route.

@iliekcomputers Which means the only way out is making those models pay for themselves.

They are releasing models you are too poor to use