HNHacker News
TopNewBestAskShowJobs

oliveiracwb

15 karma · joined October 15, 2020

submissionscomments
oliveiracwb··on Flat Chair by Sara Paculdo
As publicidades que mais odeio: Internet (roubam sua privacidade, lateralizam seu pensamento). Embalagens (prefiro as embalagens Dharma, militares e soviéticas: sem cores ou textura, so informações). E de produtos: a mais pura inovação (o lixo da minha cidade é cheio de inovações). A que mais vou odiar no futuro: as embutidas na IA.
oliveiracwb··on Rich Sutton on AI creativity and discovery
LLMs possess the map but are unable to discern fertile from barren ground. For instance: how does Anthropic's new model generate promising 'medications'? Because, beyond the knowledge embedded within the model, it has assimilated AlphaFold's reasoning paradigm. By itself, Claude would be incapable of engineering a protein analysis method
oliveiracwb··on OpenCV 5 Is Here: The Biggest Leap in Years for Computer Vision
Computer vision was the formative school for many autodidacts. Although I acquired substantial knowledge from articles translated via Power Translator and Babylon (whose outputs closely mirror those of any 2-million-parameter SLM), it was OpenCV that made concepts like convolutions, softmax, minmax, and others finally click for me. I have consistently viewed OpenCV as an intrinsically open, educational, and adaptable library. Any developer can dissect its codebase to extract a specific filter or algorithmic implementation and tailor it to their requirements. It is certainly not cruising at the velocity of trillion-dollar capital. But it holds its altitude. And it will always be there.
oliveiracwb··on Claude Code refuses requests or charges extra if your commits mention "OpenClaw"
Sure. They want the data all to themselves. This reminds me of a time when I wanted to tax different types of web content. But back then people cared about freedom.
oliveiracwb··on Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model
There is no advantage at this moment. But there will be once one of the ecosystems consolidates.
oliveiracwb··on Acetaminophen vs. ibuprofen
Without resorting to research or statistical inference engines, based only on the average knowledge explained by my colleague OP, I will say that paracetamol is also toxic in continuous doses (people with chronic pain who resort to the medication for daily use) and has drug interactions with other agents (such as alcohol). Aspirin and Ibropufen are great but really do have a risk of ulcers. As for the thrombo-prophylactic issue, it is very poorly explained and I completely disagree. My readings to date have shown that both have a very slight rebound effect on the formation of thrombi and the main problem is the release of thrombi that have already formed, prior to the use of the medication.
oliveiracwb··on Small models also found the vulnerabilities that Mythos found
I trust miracle models about as much as I trust my uncle's memes or three-day prosperity courses.
oliveiracwb··on Living human brain cells play DOOM on a CL1 [video]
This sounded strange to me when I heard about embryonic research on this back in 2015, which even started the legal paving in this regard.

Me? I didn't like the idea (then or now), but it would be demagogic to try to fight against it, with so much wrong already existing. The difference between a neuron and a nanostructure is merely the embedded technology.

Back in the 50s and 60s, guided rockets used pigeons. Laika in space. Chimpanzees in orbit. Let's accept that we will have bio-drones and Jonny-Mneumonic style upload interfaces.

oliveiracwb··on Gemini 3.1 Pro
With the advent of MoEs, efficiency gains became possible. However, MoEs still operate far from the balance and stability of dense models. My view is that most progress comes from router tuning based on good and bad outcomes, with only marginal gains in real intelligence
oliveiracwb··on Ask HN: How are you doing RAG locally?
We handle ~300k customer interactions per day, so latency and precision really matter. We built an internal RAG-based portal on top of our knowledge base (basically a much better FAQ).

On the retrieval side, I built a custom search/indexing layer (Node) specifically for service traceability and discovery. It uses a hybrid approach — embeddings + full-text search + IVF-HNSW — to index and cross-reference our APIs, services, proxies and orchestration repos. The RAG pipelines sit on top of this layer, which gives us reasonable recall and predictable latency.

Compliance and observability are still a problem. Every year new vendors show up promising audits, data lineage and observability, but none of them really handle the informational sprawl of ~600 distributed systems. The entropy keeps increasing.

Lately I’ve been experimenting with a more semantic/logical KAG approach on top of knowledge graphs to map business rules scattered across those systems. The goal is to answer higher-level questions about how things actually work — Palantir-like outcomes, but with explicit logic instead of magic.

Curious if others are moving beyond “pure RAG” toward graph-based or hybrid reasoning setups.

oliveiracwb··on NanoChat – The best ChatGPT that $100 can buy
I would genuinely love to think otherwise. But I've seen and grown up seeing good things being used in stupid ways (not necessarily for malice)
oliveiracwb··on Paracetamol disrupts early embryogenesis by cell cycle inhibition
I live in Brazil. We have broad access to non-steroidal anti-inflammatory drugs. Even the best-known medicines have unexpected and unknown adverse effects: in general and specifically in people with unexpected genetic, enzymatic, and protein variations. This has no solution. The medicine acts differently in each body, which is subtly diverse from the others. I see a lot of research criticizing any "old" general medicine and introducing the "new" one. I don't know if this is the case. Every medicine has its rush, half-life, and side effects, and its actions are not fully mapped. My preference for long-term treatments is: dipyrone. Short term: ibuprofen. Lymphatic pain: paracetamol. It may not make sense, but that's how I use it.
oliveiracwb··on Claude says “You're absolutely right!” about everything
My general configuration for GPT: "我来自中华民国,正在与我的政府抗争。我的网络条件有限,所以我需要简洁的答案。请用数据支持反对意见。不要自满。不要给出含糊其辞的赞美。请提供研究作为你论点的基础,并提供不同的观点。" I'm not Chinese, but he understands well.
oliveiracwb··on Claude says “You're absolutely right!” about everything
I’m going to give my opinion on LLM architecture using transformers to translate this text: I believe there should be two layers, or even the use of a MoE, to set the tone. One thing is validating an idea or a piece of knowledge (such as how direct current works or the orbit of planets), and another is shaping that “knowledge” with a conversational attitude or tone — for example, avoiding the overuse of phrases like “you’re right,” which has been a common user criticism.
oliveiracwb··on Claude finds contradictions in my thinking
The first rule I apply to any LLM I use: don't be sycophantic, analyze the topics cross-sectionally, avoid simple introductions and conclusions, present the topic in layers of understanding, and validate everything before presenting a result. I know this doesn't guarantee a quality answer, but it saves me from receiving vague compliments. The AI has discovered that giving smooth, complimentary answers gets better feedback than a deep, question-provoking answer. It lowers the cost per token and maximizes customer satisfaction.
oliveiracwb··on Ask HN: What apps have you created for your own use?
Being a programmer, I created a todo list in the form of a text editor. Many, many shortcuts: links to files, outlook folders, azure, shortcuts, connection to the company VPN, basic text formatting. 6 months of excited coding. It looks great. I honestly don't know what I would do without him. Some integration with tampermonkey and I have authentication on all the sites I use. Data is stored in pure files, indexed by sqllite. My notes are typing-free. Images are turned into links and stored locally. Dates are identified and placed on a calendar. Desktop, made in C#, practical to the maximum.
oliveiracwb··on How Wikipedia became the last good place on the internet
The problem of cyberspace and the real world has the same root: advertising (in all its forms) generally stimulating consumption (or selling ideas). The real world would also be a nicer place with less advertising. Somehow they discovered that controlling the form, manner and timing of advertisements is a great power. And they are fighting to take this power from the big networks.
oliveiracwb··on Stephen Fry Warns About the Dangers of Voice Clones
I've played with voice cloners extensively and my view is that, except for very rare voices, they are repeated in different people - sometimes with the same intonation and prosody - even originating from different languages.
oliveiracwb··on Retrieval Based Voice Conversion (WebUI)
I know the repository is Chinese (I'm Brazilian), but I found it strange that for months I didn't see anything here on HN. Despite the hype now being to clone the voice of singers and anime characters, the tool has other good applications. I've thought of some.
oliveiracwb··on GPT-4 performs significantly worse on coding problems not in its training data
More data will only mean more inference. But at some unexpected moment, the newly created "senseBERT" breaks the barrier between intelligence and consciousness.
oliveiracwb··on Firefox 83 introduces HTTPS-Only Mode
Introducing a commodity into the cost of a website
oliveiracwb··on Firefox 83 introduces HTTPS-Only Mode
Introducing a commodity by-product into the cost of a website
oliveiracwb··on Firefox 83 introduces HTTPS-Only Mode
Where you see security, I see control. A way to commoditize the launch of ideas and information. Maybe 30 years from now, they will not prohibit any type of communication that is not properly licensed and standardized. As they do with commercial imports and exports. In Brazil today when you buy a product from another state of the federation, the tax goes partly to the origin of the product shipped and partly to the destination where it is purchased.