kgeist

18 ms·

301.
by kgeist 1y ago
At work, we use RabbitMQ in most projects. Runs OK on a laptop.
302.
by kgeist 1y ago
I gave ChatGPT the project's README and asked it to write Go code based on it. The code it produced was very similar in structure to the original, with many parts almost identical. For example, what ChatGPT gave me: func (q *Queue)
303.
by kgeist 1y ago
>From a domain point of view, some are skeptical that bytes are adequate for modelling natural language If I remember correctly, GPT3.5's tokenizer treated Cyrillic as individual characters, and GPT3.5 was pretty good at Russian.
304.
by kgeist 1y ago
In Go, you can marshal JSON to a map, so you can extract fields manually, too.
305.
by kgeist 1y ago
>It's a real shame that the internet has driven us to assume everybody is a novice. Sorry, English is not my native language. I didn't mean to say you're a novice. >Usually, you take in that struct, and then immediatly
306.
by kgeist 1y ago
>you have a model with 20 fields and need to create a tiny slice of it (with let's say three fields), you need a few lines of boilerplate: create a record with those fields (1-3 LOC): >create a mapper for it: > ... >Go deve
307.
by kgeist 1y ago
1. As shown in the article, exposing the internal model to APIs directly is a security footgun. I've seen sites get hacked because the developer serialized internal objects without any oversight just like in the article and accidentall
308.
by kgeist 1y ago
Popov is a very common surname in Eastern Europe, something like Smith or Rodriguez. At work, we have 5 Popovs :)
309.
by kgeist 1y ago
GPT4o isn't considered an "advanced" LLM at this point. It doesn't use reasoning. I gave your prompt to o3 pro, and this is what I got without any hints: Historic shipwrecks (1850 → 1970) • ~20 000 deep water wreck
310.
by kgeist 1y ago
Indeed, they use FAISS for vector search, and the actual innovation is to store PDFs in a MP4 file. I think they should compare this to just compressing PDFs using ZIP, 7z, or RAR to see if it actually makes things better.
311.
by kgeist 1y ago
We use go-git in one of our Go projects that relies on Git for version control of its data.
312.
by kgeist 1y ago
>There are a lot of qualifiers on this: Only in Singapore, We had a big client from Singapore who only agreed to buy our SaaS subscription after we integrated SingPass (Singapore's national digital identity system) for user login. W
313.
by kgeist 1y ago
D already had this, 2 "standard" libraries: https://news.ycombinator.com/item?id=6066174 >The reason it caused the division at community is that the ones that you could not use both on the same project, this me
314.
by kgeist 1y ago
900 ms sounds like a lot for just 10,000 documents? How many chunks are there per document? Maybe Pinecone's 820 ms includes network latency plus they need to serve other users? In Go, I once implemented a naive brute-force cosine sear
315.
by kgeist 1y ago
I run Qwen3-32B with Unsloth Dynamic Quants 2.0, quantized to 4+ bits, and with the key-value cache reduced to 8-bit. It's my favorite configuration so far. This configuration has the best quality/speed ratio at this moment imho.
316.
by kgeist 1y ago
>I rewrote a simple RAG ingestion pipeline from Python to Go I also wrote a RAG pipeline in Go, using OpenSearch for hybrid search (full-text + semantic) and the OpenAI API. I reused OpenSearch because our product was already using it fo
317.
by kgeist 1y ago
I've found that Linux already does this out of the box: if you load a model once, subsequent loads are much faster because the OS caches the on-disk pages entirely in RAM (assuming you have enough RAM). If you switch to another model a
318.
by kgeist 1y ago
>Discovery of ripasudil as a novel dAMD treatment: Using the data from the first round of drug candidate testing, Robin proposed a second set of drug candidates. We tested them in the same experiment to find there was a new top hit: ripa
319.
by kgeist 1y ago
I agree it doesn't work for all cases. In our case, some services can have complex, service-specific access control logic that's hard to express declaratively in a token, so we also have to make some checks by consulting the DB. I
320.
by kgeist 1y ago
I heard new AI crawlers extract text from images too, hoping to get that extra dataset. So with that kind of blur, it might end up in LLMs.
321.
by kgeist 1y ago
>We experimented once with trying to put permissions on a JWT (more complex than your popular scopes) but that makes them grow quickly. We solved it by simply using bitmasks. Say, you want to encode an access rule "allows reading fr
322.
by kgeist 1y ago
Feels like fiddling with snapshotting raw C++ memory is a can of worms (you listed some of the pitfalls yourself). Most of my snapshotting happened at the scripting runtime level, where everything is well-defined and well-understood: I manu
323.
by kgeist 1y ago
>Many people often say that making an engine from scratch takes too long. But how long does it take to properly learn Unreal or Unity such that you can have an idea and turn it into a game without friction? Presumably, once your engine i
324.
by kgeist 1y ago
>People whose main environment is Linux intentionally buy hardware that works flawlessly with Linux. Hm, recently I bought a random "gamer PC" for the beefier GPU (mainly to experiment with local LLMs), installed Linux on it, a
325.
by kgeist 1y ago
>That's because models have training cut-off dates When I pointed out that it used a deprecated package, it agreed and even cited the correct version after which it was deprecated (way back in 2021). So it knows it's deprecated
326.
by kgeist 1y ago
Today I tried "vibe-coding" for the first time using GPT-4o and 4.1. I did it manually - just feeding compilation errors, warnings, and suggestions in a loop via the canvas interface. The file was small, around 150 lines. It didn&
327.
by kgeist 1y ago
Interesting: Russia - Europe = Putin Ukraine + Putin = Russia Putin - Stalin = Bush Stalin - purge = Lenin That means Bush = Ukraine+Putin-Europe-Lenin-purge. However, the site gives Bush -4%, second best option (best is -2%, &
328.
by kgeist 1y ago
If you want to be serious about this, use a self-hosted CI/CD server such as TeamCity, and then mirror all dependencies on a local git server (Go supports GOPROXY variable, but you can also probably fiddle with local DNS and self-signe
329.
by kgeist 1y ago
If GitHub went down, the code would be fine (just announce a new official URL), but the main thing that would be lost is issues and pull requests. Maybe Git should add official support for issues and pull requests in its metadata to be full
330.
by kgeist 1y ago
In my benchmarks for a service which is now running in production, hybrid search based on both keywords and embeddings performed the best. Sometimes you need exact keyword matches; other times, synonyms are more useful. Hybrid search combin

More ›