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
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.
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.
Oh missunderstanding here, i meant a great laptop. At the time when i got my mbpro with m1 chip there werent competitors in terms of battery/keyboard/screen and temp/noise.
So i'm looking for a laptop with equal great hardware that i can run linux on. Ofc you can run linux on great hardware, makes it even better!
NVIDIA made the 12V-2x6 port retro compatible with the previous 4000 series 12VHPWR. If you make your port compatible with past gen and it breaks with it its a design flaw, dont make it retro compatible.
This is not a user error, this is NVIDIA design error.
I think is the unreasonable ammounts that people end up paying in US that make it crazy to understand for non US.
For me as an European, i see anything related to healthcare as core to our society. You mention food, if i dont like an apple from a vendor i can just have another fruit that is cheaper this season. I wouldn't have that choice to change meds for my current ilnesses.
Yeah sure someone has to pay for it, but do meds need a 10000x markup compared to the 2x markup on food? US is such an advanced country in terms of science and tech that is scary to think people still die due to not being able to pay their cancer treatment.
eComID | https://ecomid.com | ONSITE | Stockholm, Sweden | Full-time | Machine Learning Engineer (Mid-Senior)
eComID is a Stockholm-based startup dedicated to reducing product returns in the fashion industry. We blend advanced technology with industry insights to deliver solutions that benefit retailers and customers alike. With €2.75 million in funding, we're expanding rapidly and seeking a Machine Learning Engineer to join our team.
We're looking for someone with 2+ years of experience in machine learning or software engineering with ML expertise. You should have a strong foundation in statistics, machine learning, and deep learning, and be comfortable working with technologies like Python, PyTorch, and more. Your role will include building scalable infrastructure, improving AI models, and optimizing machine learning operations.
If you're passionate about impactful work and enjoy tackling steep learning curves, we'd love to hear from you. For more details and to apply, contact us at career at ecomid dot com with the subject "Machine Learning Engineer."
So basically trained a model on minecraft. This is not generalistic at all or whatsoever. Is not like the game comes from a prompt, it probably comes from a bunch of finetuning and gigadatasets from playing minecraft.
Would love to see some work like this but with world/games coming from a prompt.
eComID | https://ecomid.com | ONSITE | Stockholm, Sweden | Full-time | Machine Learning Engineer (Mid-Senior)
eComID is a Stockholm-based startup dedicated to reducing product returns in the fashion industry. We blend advanced technology with industry insights to deliver solutions that benefit retailers and customers alike. With €2.75 million in funding, we're expanding rapidly and seeking a Machine Learning Engineer to join our team.
We're looking for someone with 2+ years of experience in machine learning or software engineering with ML expertise. You should have a strong foundation in statistics, machine learning, and deep learning, and be comfortable working with technologies like Python, PyTorch, and more. Your role will include building scalable infrastructure, improving AI models, and optimizing machine learning operations.
If you're passionate about impactful work and enjoy tackling steep learning curves, we'd love to hear from you. For more details and to apply, contact us at career at ecomid dot com with the subject "Machine Learning Engineer."
This research isnt very novel either. Transformers for sequential event handling and recommendations is a known thing. Pretty sure this is what spotify uses on their autocomplete shuffle playlist (fill in songs between your songs, similar to how autocompleting text would be)