Browse Comments — Clean (de-noised)

hey the biggest leap we need in ai right now isnt just better capabilities its moving from controlling model behavior to real consequence controlonce these systems can reason across modalities and act in the real world we have to ask the hard questions what evidence drove the decision what actually changed can it be reversed and most importantly who is accountable when something goes wrongsafety cant stay abstract it needs to become operational with clear auditability and human oversight especially as we push into agents and roboticsthis controllability gap feels like the next major bottleneck what do you think is the most important layer we should be building right now to make ai truly safe in the wild

The workflow-fatigue point is real, but I’d be careful with the “one workspace for everything” conclusion. Centralizing tools can reduce friction, but it can also blur the reason different models are valuable in the first place. Different models have different strengths: reasoning, coding, UX, retrieval, image generation, long-context analysis, structured outputs, or synthesis. If everything is pushed through one workspace, the risk is that orchestration becomes easier while evaluation gets weaker. The hard part is not just accessing multiple models. It is knowing which model should handle which step, how outputs should be compared, and how QC happens across the workflow. The future is not only “one platform.” It is governed orchestration: routing the right task to the right model, then validating the output before it becomes execution.

The LLM’s won’t have anything to train on. Hmmm...

Well said—declaring "RIP ChatGPT" misses the mark, like swapping out a screwdriver and calling carpentry dead. The real magic is how teams stitch together these AI models into cohesive, value-generating systems that actually do the work, not just talk about it. That’s exactly where platforms like shine. Instead of chasing the latest model hype, Chat Data lets you design robust workflows, link multiple agents, and automate complex processes across text, voice, and much more. It’s about building a reliable AI-powered machine shop, not just showing off shiny new tools.

Cool...

Fantastic explanation here!

麻煩幫我跟您們家執行長說:「所謂的物理晶片範疇,皆隸屬於“超晶體-超導體架構認知範疇”,這部份學術認知重點在於「粒子學」與「相對論」的應用重點,也就是智能發展的主要架構資訊體系」,超晶體一直以來都是我在追蹤與關注的分析重點事務!

Brilliant fantastic blessings 🙏

OpenAI’s new reasoning model produced a novel proof disproving a long‐standing Erdős geometry conjecture, marking what the company calls the first autonomous AI solution to a major open math problem

$2000 per engineer per month in tokens and the AI still ask for clarification on requirements. The human at least just guessed and moved on.

One thing I find interesting is that AI is making proof of work far more valuable than credentials. When everyone can generate output, the differentiator becomes showing how you think, what you build, and the problems you consistently solve. That's where documenting publicly starts compounding.

Agree. Using a single model is great for the people who just want to play. But those that are actually using AI, aren’t just using a model, they are building an ecosystem with the LLM just being a small part of it.

Stack

We all in one thing is inevitable, but I don’t understand how anyone can access all those platforms without paying for them

Thamk Nk. For the thing

Alex Smirnoff To my understanding, France and Russia both have ER-oriented healthcare cultures (urgences in France, skoraya pomoshch in Russia), so this is aligned with the US.

The interesting shift is that engineering leverage is increasingly coming from coordination and system design, not just the amount of code one person can write.

The real unlock is not only faster answers. It is lower friction around exploration. In business, the same idea applies when AI helps teams test options, preserve reasoning, and move from a rough question to a better decision without losing the trail.

AI doesn’t replace ambitious people.It exposes passive ones. The students winning right now aren’t always the smartest in the room. They’re the ones shipping weird projects at 2AM, learning in public, breaking tools, fixing them, and moving faster than curriculum updates. Meanwhile some executives are still scheduling a 6-week meeting cycle to discuss whether AI is “relevant.”By the time the committee approves the pilot, an intern has already automated half the workflow with three prompts and a coffee. The future probably belongs to people who can do both:think clearly like humansand move insanely fast with AI.

This is great research. Thank you for doing this.