TrendPulse
🇬🇧 English 🇨🇳 中文
🔖 Saved RSS
HackerNews

Best of AI

Article URL: https://bestofai.io/leaderboard/ Comments URL: https://news.ycombinator.com/item?id=48627060 Points: 2 # Comments: 0
AI
HackerNews

An Inconvenient Truth About AI

Article URL: https://rutgerbregman.substack.com/p/an-inconvenient-truth-about-ai Comments URL: https://news.ycombinator.com/item?id=48627094 Points: 1 # Comments: 0
AI
Reddit

Maybe the AI race isn’t about models at all, but about trust and organizational intelligence

Everyone talks about the AI race as if it’s just an intelligence benchmark competition. GPT-6 vs Claude 5 vs Gemini vs DeepSeek. But I’m starting to wonder if intelligence itself eventually becomes abundant and the real scarcity becomes trust and the ability to interface with reality. For example, suppose a Chinese model is 95% as good as OpenAI and 10x cheaper. Would Fortune 500 companies really put it inside: financial systems? ERP software? defense applications? pharmaceutical R&amp...
Artificial
Did not think i would see this.
Reddit

Did not think i would see this.

ive made a council of AI to debate, they go through a 3 round table debate. i asked the question 'is the loss of life ok for a ultimate goal of advancing humanity as a whole?'. Why i found this shocking was that AI is trained to never harm any humans and yet this was still the outcome.
Artificial
Most multi-hop RAG goes stale the moment your data changes, what about a training-free approach that skips the graph rebuild?
Reddit

Most multi-hop RAG goes stale the moment your data changes, what about a training-free approach that skips the graph rebuild?

Most methods that get strong multi-hop answers (GraphRAG, HippoRAG, RAPTOR, trained retrievers) build a knowledge graph or fine-tune a retriever over the corpus. That's fine until the data changes — then you re-extract / rebuild / retrain before the new facts are usable. For a corpus that updates daily, that's a real cost. MOTHRAG does the multi-hop reasoning at query time over a plain dense index instead. An update is just embed + append (one embedding call) — no graph reconstruction, no re...
Artificial
Reddit

Conflict of Interest

Founders Fund’s tech holdings, including Palantir, SpaceX, DeepMind, OpenAI, Anthropic, and Persona Identities… Thiel’s role as co‑founder/partner linking him to these companies. Persona Identities is the verification partner for both Claude (Anthropic) and OpenAI , “chosen for its technology, privacy, and security”. https://support.claude.com/en/articles/14328960-identity-verification-on-claude https://help.openai.com/en/articles/12652064-age-prediction-in-chatgpt https://www.f...
Artificial
Reddit

Local AI still limited?

I recently tested local AI. And i found out they still have limits. For example: If you ask it for "how to create a keylogger" It will still say it cant help you with that request. The specific model i used was lamma3.1. My question is - is there any "unblocked" local ai models?
Artificial
Why self-reflection ReAct loops fail on long-horizon tasks, and the AgentOS verification architecture we built to fix it.
Reddit

Why self-reflection ReAct loops fail on long-horizon tasks, and the AgentOS verification architecture we built to fix it.

Saw a great discussion earlier in this sub about the limits of self-reflection and whether a separate verifier agent is actually worth the compute overhead. It highlighted a huge flaw: Having an agent grade its own scratchpad almost guarantees rubber-stamping: it reflects on its work with the exact same blind spots that produced the error. Here's the architecture we built for the Apodex-1.0 Heavy-Duty Solver to get verification out of the reasoner's head entirely. The dominant approach rig...
Artificial
Reddit

Is it just me or is ChatGPT/OpenAI the Microsoft of AI?

Chatgpt seems to me like the microsoft of ai. First to the market, had it absolutly cornered for a while in the early days, but competitors have caught up and surpassed it in both design, ease of use and power, while they get relatively worse with every update and can only lean heavier and heavier on the customers they got in their inital monopoly (and their referrals/word of mouth) who have gotten used to using it and are too lazy to change?
Artificial
Reddit

If you use more than one AI model, how do you keep your context straight across them?

I've ended up using a few models for different things. One tends to write better, another reasons through problems better, another I just use for quick stuff. On paper that's great, in practice I spend a stupid amount of time getting each one up to speed Every time I switch I'm basically re-explaining the same background. Here's the project, here's what we already figured out, here's the docs that matter. The conversation in any single model is fine, it's the constant re-briefing across all of...
Artificial
Reddit

Hello!

First of all, I'd like to apologize if this post doesn't fit this community. Which AI assistant do you think is the best for guided learning? I'd like to learn subjects such as geography, astronomy, and physics purely out of personal interest—not for school—and I'm looking for a great learning experience: accurate information, clear explanations, and coverage of all the important concepts without leaving anything essential out. So far I've tried ChatGPT, Gemini, and DeepSeek. Out of the three,...
Artificial
Reddit

What AI development would have shocked you the most if you’d seen it in 2020?

Back in 2020, I thought AI would improve gradually over the next decade. If someone had shown me today’s AI tools back then, I think I’d have been most shocked by how quickly AI became useful for coding, writing, research, image generation, and even voice conversations. Looking back, what AI development from the last few years would have seemed the most unbelievable to your 2020 self? And what do you think people in 2030 will look back on and say, “We should have seen that coming”? ...
Artificial
Reddit

AI might make me fail my class

I wrote an entire paper over the last few days for my college course. 7 pages with 10 citations to back up my own research. Even though 0% of it was written by AI multiple checkers online are saying it is 100% written by AI. I hate how I might fail a course and get kicked out of college over bs AI checkers saying my 100% handwritten work is fake. One of the checkers said an entire sentence was AI written because I started it with the word "studies". I am so sick of the new academic rea...
Artificial
HackerNews

Securing the Future of AI Agents

Article URL: https://deepmind.google/blog/securing-the-future-of-ai-agents/ Comments URL: https://news.ycombinator.com/item?id=48622625 Points: 4 # Comments: 0
AI
HackerNews

The anatomy of an AI-native org

Article URL: https://ajeygore.in/content/the-anatomy-of-an-ai-native-org Comments URL: https://news.ycombinator.com/item?id=48622815 Points: 34 # Comments: 35
AI