jiwidi

always one in the room

Yeah, and seems like the automated systems have been failing a lot recently. Or that they didnt correctly designed them for this case.

This means the health of those systems and mantinability by whoever was responsible hasn't been well done.

I agree with you, once the only thing and death is a human, nothing can be trusted, humans will make mistakes.

Exactly, this can't happen. You need to put measures on this.

Both the maintenance issue and bad systems that lead to human error creating this crash need to be taken responsability by the public system and fix them.

Another one.

The public minister of transport is still on his chair after the last ones. No responsability admited and blaming the far right for attacking him with media.

Is amazing how the current politic scene at spain has shittified, they get away with everything and ppl will still clap for them.

Twice? uiuiui

Spent the weekend turning small LLMs into decision models, with some good results and a lot of learnings along the way. Topping the decision index leaderbard for their categories https://huggingface.co/spaces/multimodalart/jev-decision-ind... Sharing a 0.8B model, when evaluated on the decision Index it tops the sub-1B category, 45.7% above the best other Qwen3.5-0.8B fine-tune on the leaderboard.

Most of work was data calibration, finding external datasets and readapting them towards this scenario, so i expect a lot of improvements and work like this to come from community and push these numbers even higher. Even more if we get qwen 3.8 releases for these model categories.

I wonder, how different is this now from qwen trained rerankers like zeroentropy ?

big agree here.

People dont seem to realize prompt to code is just another interface. Is the human review part that we should care about, did this code get reviewed and reiterated on? Is it fully AI generated and reviewed?

Those are the important tasks.

Whoever feels like a better programmer because they write all the terraform files by hand and not use an llm, dislike.

My mistake for calling it "copy", should have said fork maybe (?)

They are awful close to still require a reference to the original project imo. They are achieving the same output (identify birds and display them in a frame display all hosted in a raspberry pi) with that twist to how to generate the art.

That's fair, and I totally get wanting to give it your own spin.

I still think it would be good to acknowledge the original project that sparked the idea, especially now that you've mentioned it was one of the inspirations.

Definitely keep going with yours and evolve it further. That's how good projects grow. But for a healthy open source community, I think we should make a point of crediting the people whose work inspires us. Otherwise, over time, it can discourage people from sharing their projects publicly in the first place.

Yeah agree

I understand OP is continuing the work and improving it as he sees fit, but that in my mind is a fork and should aknowledge the initial work that sparked it.

Maybe i'm wrong tho and this is completely parallel work! Waiting for OPs reply

Its awful close with no attribution or mention of the original project, which should be the minimum i feel like.

Isnt this a fork* of https://x.com/WarnerTeddy/status/2060018688645115964?

https://github.com/Twarner491/AvianVisitors

https://theodore.net/projects/AvianVisitors/

Or maybe im missing to see is the same author? It went viral a few months ago

what, so you go and bomb them then now u pay them hahaha

US is the world best at american football :fireworks:

See original opus 4.6 sitting at 16% hallucination and the retest on 12th of april at 33%

They definitely must be doing some quantization or optimization to meet demand, otherwise why would model performance degrade this much? It's been crazy for me personally

Hi Boris,

Long term claude code user here. Is the first time i've had to setup a hook to codex to review claude output.

Is hallucinating like never before

Is missing key concepts/instructions in context like never before

Is writing bad code that will "pass test" much more. Before it use to try be critic and do good code, now it will try to hack test and bypass intructions for a green pass.

tell me what pc with an nvidia gpu can you buy with same memory and performance.

I never liked apple hardware, but they are now untouchable since their shift to own sillicon for home hardware.

Great potential! Love the idea

suddenly when war comes is only mandatory for man

Very interesting, how do you do that? Do you limit yourself what you feed or via custom instructions? I had a similar case so would love how you are doing the prompting here.

In my case we went with embeddings and clustering to find close papers to each other because llm were allucinating.

Pretty cool! How do you do to build these "stories" based on news?

thats go for you my man