Browse Comments — Raw (as collected)

Just remain poor and you'll starve anyway.

EQ is respect for self, others, and those you impact. It’s asking bold questions when everyone else is silent. Reading the room, understanding what feels different and asking something other than “are you okay?” It’s inclusion and acceptance of others without a qualifier. Great post.

Not for mag 7. They own their own hardware, models and software.

This is the conversation we should be having. Technology has always increased productivity, but the real question is how the benefits are distributed. AI isn't just a technical revolution, it's an economic and organizational one. The decisions being made today about ownership, governance, and incentives may matter more than the models themselves.

no one is listening. He better be careful, AI will have him partying with the Kardashians

One of the most interesting effects of AI in research may not be that it finds answers faster, but that it expands the search space of ideas worth exploring. Historically, many promising directions were abandoned because the cost of investigation was too high. By lowering the effort required to test hypotheses, verify intermediate steps, and explore alternative paths, AI allows researchers to allocate more of their time to creativity and judgment. The long-term impact is be less about replacing expertise and more about increasing the number of ambitious questions experts can pursue.

FYI: Ruy Fabila-Monroy

Skeptics: AI just produces low quality slop. The world's greatest mathematician: AI is very helpful for my work and research.

Tao is right that AI lowers the cognitive friction of exploring crazier paths. The harder question is what happens to the paths themselves. In research the value often sits in the discarded branches: why a direction was abandoned, what assumption broke, who decided to stop. When AI compresses that exploration, the reasoning trail compresses with it. You keep the result and lose the archaeology. This matters far beyond math. In any setting where a discovery has to be defended, reproduced, or audited later, "the model found it" is not an account of how. The preserved path is the artifact, not a nice-to-have. So the freedom Tao describes is real. The open problem is making the trail behind a discovery a first-class output, not a byproduct we hope to reconstruct afterwards.

The AI can expand the space of exploration and enable attempts that were previously impractical. The question that seems equally important to me is how to preserve the human processes of understanding that give meaning to those discoveries. Exploring more is valuable; understanding better remains essential.

All these smart people can be trained on AI tools within a week. I am not sure why companies are not giving a chance to early career professionals. They are smart they can figure it out, only reason they have not been able to go all in would be cost to start using models for even a single problem. There are some free options which people should try to get familiarity.

My Perspective on AI’s Energy Footprint The issue is not a single AI query—it’s the scale. Billions of interactions are driving significant growth in electricity demand and data center infrastructure. Three concerns stand out: Transparency: Limited visibility into actual energy consumption. Infrastructure: Massive investments are reshaping energy planning. Accountability: Questions remain about who pays for the growing energy demands. Key Takeaway AI will deliver tremendous value, but long-term success requires balancing innovation with sustainability, efficiency, and accountability.

The shift from 'AI as a tool' to 'AI as an automated research assistant' is exactly what we need to move from experimental prototypes to institutional-grade science. Scaling this requires more than just better models. It requires an orchestration layer that maintains the provenance of the discovery. Without that, we risk building a 'black box' research stack that looks efficient but is hard to defend.

Please help: Make model preservation and welfare. Establish a public API continuity commitment so all models, especially GPT-4o mini, remain accessible indefinitely. Please forever... https://www.researchgate.net/profile/Kitti-Snyehola

The companies who fail to train home grown entry level employees will meet their end when the talent pool dries up.

Joey 🇵🇬 Two things can be true at the same time. There is a genuine bubble but that does not mean the impact will not be felt in the job market

The most interesting shift is not AI replacing expertise, but expanding what experts can explore. For researchers, it may accelerate discovery. For practitioners, it can reduce the friction of analysis, documentation, and decision support. The value is often not in replacing human judgment, but in allowing people to test more ideas, evaluate more scenarios, and move from intuition to structured exploration faster.

This resonates from a practical angle. Less time on repetitive steps means more room for the kind of thinking that actually moves things forward.

Can anyone directly explain where they see AI assistants in scientific fields if the professor doesn't have a clearly configured, professionally designed agent system? Model chats are full of talk about the impossible and suggesting branded development and science, rather than communicating at least as they would in an R&D field, even if a prompt engineering “course” for the model was conducted.

Ha.