HNHacker News
TopNewBestAskShowJobs

dmezzetti

861 karma · joined May 22, 2021

Founder of NeuML (https://neuml.com)

Creator of txtai (https://github.com/neuml/txtai)

submissionscomments
dmezzetti··on Indie, alone, and figuring it out
Sounds about right for someone who is building on their own.
dmezzetti··on The current state of the theory that GPL propagates to AI models
I've spent years building something for free with no expectation of anything in return. Perhaps someone just doesn't believe in the GPL and has no ulterior motive for that.
dmezzetti··on The current state of the theory that GPL propagates to AI models
Once it's out of my hands, so be it. The users can choose not to use derivatives.
dmezzetti··on The current state of the theory that GPL propagates to AI models
I think it makes a lot of sense for hobby software and non-commercial software. It's just tough to do in a commercial setting for a number of reasons.

So ultimately while good intentioned, you end up limiting how many people can use what you've built.

dmezzetti··on Linux Kernel Explorer
Something as simple as an LLM prompt that explains what the file does would go a long way I think.
dmezzetti··on Age of "Don't do it yourself"
I agree. If everyone always just used the standard solution we'd never get better ones. Nothing is ever done.
dmezzetti··on Ask HN: Why do people say LLMs create bad code "quality"?
Good luck!
dmezzetti··on Ask HN: Why do people say LLMs create bad code "quality"?
The perception is that it's sloppier than what humans write. I wouldn't waste too much time worrying about that if the code is solving your problem. Sounds like you need to focus on getting a MVP to show to potential customers and if LLMs help you get there faster, so be it.
dmezzetti··on Age of "Don't do it yourself"
In some cases, there is risk to doing it yourself (rolling your own authentication scheme). But from a pure learning standpoint, yes people should be encouraged to build even if solutions exist. Perhaps there is a better version we don't know about until it's built.

When it comes to a startup or company project though, it also depends on how much time you spend on something that isn't the core problem.

dmezzetti··on We're losing our voice to LLMs
Let's not forget to mention the rise of AI-generated video. You can't really trust any video as real anymore.
dmezzetti··on The current state of the theory that GPL propagates to AI models
Fair enough. You'd like to hope that there is a voluntary "pay it back and forward" mentality. But I understand that is a leap of faith with a lot of blind trust.
dmezzetti··on The current state of the theory that GPL propagates to AI models
I understand people want to control what happens once they build something. Too often do you see startups go with a permissive model only to go to a more restrictive model once something like that happens. Then it ends up upsetting a lot of people.

I'm of the opinion that what I build, I'm willing to share it and let others use it as they see fit even if it's not to my advantage.

dmezzetti··on The current state of the theory that GPL propagates to AI models
I'm not in anyone's business just sharing my opinion on GPL. I understand why people go GPL / AGPL just not for me. To each their own if they want to go down that path.
dmezzetti··on Linux Kernel Explorer
Very interesting! I was kind of expecting that this was going to use LLMs / coding agents to explain the kernel as you stepped through it. But this is a nice app.
dmezzetti··on The current state of the theory that GPL propagates to AI models
That's your prerogative. It's just not for me and GPL is basically something I avoid when possible.
dmezzetti··on The current state of the theory that GPL propagates to AI models
Ok but I just have no interest in imposing restrictions on how people distribute what I build in such a manner either. That's just me.
dmezzetti··on API that auto-routes to the cheapest AI provider (OpenAI/Anthropic/Gemini)
A free router is https://huggingface.co/models
dmezzetti··on Investors expect AI use to soar. That's not happening
AI can do some good things with the right expectations and people.
dmezzetti··on Original Superman comic becomes the highest-priced comic book ever sold
This is the epitome of why people go to yard sales
dmezzetti··on Cloudflare Global Network experiencing issues
This one is bigger than the AWS East outage...
dmezzetti··on The Case Against PGVector
You can make it even simpler and not bother with any of this. With even something as large as 100M vectors, you can just use Torch or GGUF with compression. Even NumPy can take you a long way. Example below.

https://github.com/neuml/txtai/blob/master/examples/78_Acces...

dmezzetti··on Ultra efficient vector extension for SQLite
Odd licensing strategy here. It's like someone that wants the cachet of saying they are open source without being it.
dmezzetti··on The Theoretical Limitations of Embedding-Based Retrieval
This paper has been misrepresented many times. At then end it says:

Multi-vector models

Multi-vector models are more expressive through the use of multiple vectors per sequence combined with the MaxSim operator [Khattab and Zaharia, 2020]. These models show promise on the LIMIT dataset, with scores greatly above the single-vector models despite using a smaller backbone (ModernBERT, Warner et al. [2024]). However, these models are not generally used for instruction-following or reasoning-based tasks, leaving it an open question to how well multi-vector techniques will transfer to these more advanced tasks.

Sparse models

Sparse models (both lexical and neural versions) can be thought of as single vector models but with very high dimensionality. This dimensionality helps BM25 avoid the problems of the neural embedding models as seen in Figure 3. Since the of their vectors is high, they can scale to many more combinations than their dense vector counterparts. However, it is less clear how to apply sparse models to instruction-following and reasoning-based tasks where there is no lexical or even paraphrase-like overlap. We leave this direction to future work.

In other words, it says that both multi-vector (i.e. late interaction) and sparse models hold promise.

dmezzetti··on Wikipedia as a Graph
I did something similar to this except of using hyperlinks, the links were based on the vector similarity between article abstracts.

https://github.com/neuml/txtai/blob/master/examples/58_Advan...

dmezzetti··on Show HN: PageIndex – Vectorless RAG
Yep, context, however you build it.
dmezzetti··on Show HN: PageIndex – Vectorless RAG
You got it!
dmezzetti··on Updates to Consumer Terms and Privacy Policy
If you don't like this, use local AI.
dmezzetti··on Show HN: PageIndex – Vectorless RAG
Context and prompt engineering is the most important of AI, hands down.

There are plenty of lightweight retrieval options that don't require a separate vector database (I'm the author of txtai [https://github.com/neuml/txtai], which is one of them).

It can be as simple this in Python: you pass an index operation a data generator and save the index to a local folder. Then use that for RAG.

dmezzetti··on Agent-C: a 4KB AI agent
Why not just do this with a shell script? It's just a wrapper around curl.
dmezzetti··on GPTs and Feeling Left Behind
A lot of this is classic gaslighting. Gaslighting is the manipulation of someone into questioning their perception of reality (per Wikipedia).

Basically, a lot of people who are experts are being told this story and they think they are the only one who doesn't get it.

There are plenty of gains to be had with AI/LLMs but just not in the way it's typically marketed.

← PreviousPage 2 of 12Next →