Browse Comments — Clean (de-noised)
Close reading of the corpus at each pipeline stage: raw → clean → relevant → coded.
4.2K
comments matched
· page 186 of 210
Leif Rydenfalk sounds like a great plan
Corneliu M.
Ryan Watters, ELS Respectfully, planned data centre construction is exactly the evidence for the claim. If trillions of dollars are being directed towards new generation capacity, transmission infrastructure, data centres, cooling systems and semiconductor supply chains, it’s because future compute demand is expected to be materially larger than today’s. The debate isn’t whether AI demand grows. The debate is whether efficiency gains can outrun demand growth. History suggests the opposite. Every major reduction in the cost of compute has resulted in more compute consumption, not less. Which is why today’s energy footprint is likely the smallest we’ll ever see.
Demis Hassabis Why are you guys (every top tech) even considering to make AGI, despite being aware of the risks of this technology?? Nobody have ever given the exact definition of AGI, so would it be Google's AGI or OpenAI's.
How to debug on your own is always valuable skill even in AI era. what if these free AI tools such as chatgpt , gemini, perplexity , claude etc suddenly turn into subscription model even for basic usage? Or some limited quota for free use? It is ok to use AI as force multiplier
劉晉丞 Everything is already done, for you it'll take 2(3) clicks to stop all the wars: 1. Subscribe to The Imagine project X account now 2. Support The Digital Strike on February 24 *3(optional). Share to speed up the end of 'wars era' P.S. We only need a support of 51% of humanity to make it work.
"What Terence Tao is describing isn't AI doing the mathematics; it's AI handling enough of the surrounding work that the researcher can stay in the problem longer. That distinction matters. Augmentation at that level looks very different from automation, and it's probably where AI creates the most durable value: not replacing the expert, but extending how long they can operate at full capacity."
Strong point. The conversation has shifted from "which model is best?" to "which workflow creates the least friction?" In my experience, the biggest productivity gains come from reducing context switching rather than chasing marginal model improvements. Curious to hear your take: which integrated workflow delivers the highest ROI for most users today? Would be great to see others share what's still forcing them to jump between tools.
AI was never cheap labour, the realization just took a year to arrive — and a year of layoffs, budget burns, and cancelled licenses was apparently what it cost to find out.
Sakib Ziad Thanks! "AI is just a Tool", is my Belief. It just frames the resultant data. The Probability brings out the structure. But AI can not think. It is after all an Idiot Model. But good to be supportive in relating the data and structuring the result in human language. A Cross thinking which we call as Human Stimulus, AI model cannot do it. We need to appreciate Discussion Forums like Stack-Overflow.
Demis Hassabis Meanwhile developers: "Cool, my AI agent just opened 47 tabs, wrote 12 microservices, and created 3 new bugs to stay employed." 🤖🔥 We’re not replacing engineers—we’re turning them into AI orchestration managers. For builders learning Agentic AI the practical way: 🚀
The pace of innovation is accelerating faster than ever. As AI capabilities continue to expand, the competition for exceptional AI, ML, and software engineering talent will become even more intense. An exciting era for builders and innovators.
There is no other blockchain, only DFINITY Foundation and Caffeine can do this.
Absolutely amazing technology.
That music in the background though.... once it clicks.... this video is pretty sadistic, actually, if you think about it. Cannibalism.... hence the dark background music.
This nails why "confidence" is the wrong metric for AI. Humans signal uncertainty with hesitation; models don't have that tell. The skill we're all quietly developing now isn't prompting it's knowing when to distrust a fluent answer.
One of AI's most impressive skills is delivering a completely wrong answer with the confidence of someone who already has a TED Talk scheduled about it. 😄 My favorite cases are when it invents a fact, doubles down on it, and then smoothly transitions into life advice. “The answer is incorrect, but have you considered mindfulness and a healthy work-life balance?” That's when you know the conversation has truly evolved.
Wrong formula. Confident delivery. And when I objected, a calm reframe that maybe I was the problem. The AI is in its situationship era and I walked right into it.
It’s worth noting something about probabilistic intelligence that confidence is not a proxy for correctness. A good skill going forward is knowing when the system is guessing confidently versus actually grounded in something verifiable.
Summarising historical information based on meeting material and minutes and then concluding what decision was made in a meeting happening 2.weeks into the future and when challenged providing arguments as to why it would be the right decision. Not that it was predicting decisions in the future. Use AI but don't blindly rely on it....its good but not yet excellent.