We're in the same space, and we knew about those shady practices for a while now, so it feels pretty good to see them finally exposed. It doesn't give back the capital they sucked in though
We were using gpt-4o for our chat agent, and after some experiments I think we'll move to flash 2.0. Faster, cheaper and a bit more reliable even.
I also experimented with the experimental thinking version, and there a single node architecture seemed to work well enough (instead of multiple specialised sub agents nodes). It did better than deepseek actually. Now I'm waiting for the official release before spending more time on it.
How many of them are even real though? I'm pretty sure Musk has a troll farm for a long time now, back in twitter days his supporters' profiles already looked very suspicious
Grok has absolutely no safety mecanism in place, you can use it for anything it will not block a query, all under the pretext of "free speech". And it's not open source either
> "It's time for OpenAI to return to the open-source, safety-focused force for good it once was," Musk said in the press release. "We will make sure that happens.""
From the creator of Grok, this is such an insane thing to say
I would even say that he's now consistently lying to get to what he wants. What started as "building hype" to raise more and have more chances actually delivering on wild promises became lying systematically for big and small things..
you would probably first tag the papers (AI papers then being tagged with "AI"),then generate for instance an Elasticsearch query with an LLM for matching papers with the ai tag and the word transformer
Yes I had a look at it, I haven't tried e5-mistral-7b-instruct yet but I'll definitely give it a go.
Is there such a leaderboard only focused on retrieval by any chance? I haven't found one so far
This is a problem we're hitting as well. Hyde (Hypothetical Document Embeddings) is a zero-shot approach to it, where from the query you generate documents that would have a very small distance to the actual documents you're looking for. For question answering this would mean generating hypothetical answers. In some cases though, it can yield better results to generate hypothetical queries for each documents, question answering actually being one of them. That requires a lot of pre-processing though, which is not always possible.
I'd be curious to hear what people think is state of the art of this kind of problem?
I think the Cohere embeddings model v3 is very good, as it handles queries and documents differently to embed them in the same vector space. Otherwise for a specific use case I guess dense retrieval (which is basically a problem specific fine tuned version of this approach) is the best way to go about it?