Hacker News: flornt

From Semantic Ablation to Attentional Ablation

flornt — Thu, 02 Apr 2026 19:06:08 +0000

Article URL: https://reflexions.florianernotte.be/en/post/ablation-attentionnelle/

Comments URL: https://news.ycombinator.com/item?id=47618784

Points: 2

# Comments: 1

New comment by flornt in "Writing with LLM is not a shame"

flornt — Sun, 24 Aug 2025 11:04:32 +0000

Thanks. Really appreciate your comments. It opens some perspectives I haven't considered and gives more things to think about regarding this. I'll digest it and update the content based on your observations!

Writing with LLM is not a shame

flornt — Sun, 24 Aug 2025 10:10:29 +0000

Article URL: https://reflexions.florianernotte.be/post/ai-transparency/

Comments URL: https://news.ycombinator.com/item?id=45002958

Points: 107

# Comments: 147

From Overwhelmed to on Top: How Managers at Upflow Get Things Done

flornt — Mon, 28 Jul 2025 07:03:06 +0000

Article URL: https://upflow.io/blog/inside-upflow/how-managers-at-upflow-get-things-done

Comments URL: https://news.ycombinator.com/item?id=44708035

Points: 1

# Comments: 0

New comment by flornt in "Read-it-later app Pocket is shutting down – here are the best alternatives"

flornt — Fri, 30 May 2025 04:34:36 +0000

I really regret Omnivore....

New comment by flornt in "Chomsky on what ChatGPT is good for (2023)"

flornt — Sun, 25 May 2025 18:09:02 +0000

I guess it's because LLM does not understand the meaning as you understand what you read or thought. LLMs are machines that modulate hierarchical positions, ordering the placement of a-signifying sign without a clue of the meaning of what they ordered (that's why machine can hallucinate :they don't have a sense of what they express)

New comment by flornt in "Ask HN: Would you use a tool that builds AI agents from a prompt?"

flornt — Sun, 18 May 2025 12:59:15 +0000

Sure. And I'am pretty much convinced that we'll have more and more tools to build things from plain language. These tools will have some extended questions to grasp what really are the user's intentions.

LLMs are more persuasive than incentivized human persuaders

flornt — Sat, 17 May 2025 20:05:09 +0000

Article URL: https://arxiv.org/abs/2505.09662

Comments URL: https://news.ycombinator.com/item?id=44016621

Points: 140

# Comments: 116

Asking a LLM for help is fine

flornt — Thu, 15 May 2025 06:37:06 +0000

Article URL: https://reflexions.florianernotte.be/post/asking-llm/

Comments URL: https://news.ycombinator.com/item?id=43992414

Points: 1

# Comments: 0

Republicans push for a decadelong ban on states regulating AI

flornt — Wed, 14 May 2025 17:13:11 +0000

Article URL: https://www.theverge.com/news/666288/republican-ai-state-regulation-ban-10-years

Comments URL: https://news.ycombinator.com/item?id=43986928

Points: 13

# Comments: 13

New comment by flornt in "Ask HN: If 1 person can control 10 AI agents, why would still need that person?"

flornt — Mon, 12 May 2025 07:38:50 +0000

see my answer below.I guess it's not quite the same thing. Self-service checkouts are just a way to buy your groceries — same with online shops. In both cases, the human is (or was) mainly there to help sell a product, not to deliver a service in the proper sense. Sure, you could argue that it’s still a form of service, but the core business of a company like Walmart isn’t about providing human services — it’s about selling goods.

New comment by flornt in "Ask HN: If 1 person can control 10 AI agents, why would still need that person?"

flornt — Mon, 12 May 2025 07:35:32 +0000

What you’re referring to is the difference between automation — following the rules — and autonomy — having the ability to break or adapt them. A traditional machine just follows instructions. It doesn’t make decisions on its own. But from what I understand, AI agents can show a certain level of autonomy. They don’t just execute predefined steps; they can adjust, prioritize, even improvise within a given framework.

New comment by flornt in "Ask HN: If 1 person can control 10 AI agents, why would still need that person?"

flornt — Mon, 12 May 2025 07:33:47 +0000

Sure, but once the solution is built, why would a company keep paying someone externally just to watch/control it? Won’t the future be about having an AI agent that can be fine-tuned to your company’s needs through natural language?"

New comment by flornt in "Ask HN: If 1 person can control 10 AI agents, why would still need that person?"

flornt — Mon, 12 May 2025 07:31:47 +0000

Yes, but it's not exactly the same. Self-service checkouts are just a way to buy your groceries — same with online shops. In those cases, the human is simply involved in selling a product, not delivering a service in itself. What I'm thinking about is an AI agent that actually provides services the way a human would — like doing strategic monitoring, replying to emails, or producing documents as part of an AI-enabled workflow."

Ask HN: If 1 person can control 10 AI agents, why would still need that person?

flornt — Sat, 10 May 2025 18:09:15 +0000

With agentic frameworks becoming more accessible, it's plausible that one skilled individual could coordinate multiple AI agents to do the work of an entire team. Ten agents handling design, code, legal reviews, content, ops—coordinated by a single human or whatever you can think.

It sounds like augmentation. But taken further, it’s also clear that this compresses the chain of value. Why pay for ten salaries—or even one intermediary—when a client could learn to command the agents directly?

Some say, “You still need a conductor for the orchestra.” That one person is the integrator, the pilot, the one who makes sure the agents are working toward a coherent outcome. Fair enough.

But here’s the counterpoint: clients already do this kind of integration. They don’t ask their suppliers to "be creative," they give precise goals and constraints. If the agents become intuitive enough to understand those same inputs directly, what’s left for the human in the middle to add?

If I used to hire a designer, and now I just prompt a design agent myself… and if the same logic applies to coding, marketing, or legal reasoning… then how far are we really from businesses where clients orchestrate their own AI “staff” directly?

Is the human coordinator role just a temporary bridge?

I’m genuinely curious: have you started working this way? Are you seeing clients question your place in the loop? What do you think we’re underestimating—or overestimating?

Comments URL: https://news.ycombinator.com/item?id=43947659

Points: 4

# Comments: 22