News for hackers. Stuff that matters.

archiv

archiv on Hacker News

Re:MuJoco Playground on Thursday January 16, 2025 @05:10PM (#42731551)
This is great - I haven't looked at the documentation yet but do you support soft body simulation?

Show HN: Polaris – Training On-Device AI Agents with Real User Data

Posted by archiv on Monday October 28, 2024 @12:38PM
from the show-hn dept.

https://github.com/cyrilzakka/Polaris

I built Polaris to bridge the gap between AI models and real-world user interactions on macOS. It's a Swift toolbox that lets you collect data (screen capture, keystrokes, mouse movements) through official Apple APIs for training and deploying macOS AI agents, with the ability to control your computer programmatically. Still a WIP but looking for feedback!

( Read More... | no comments | 2 points | on HN )
Re:Show HN: A macOS Client for HuggingFace Chat on Wednesday October 23, 2024 @04:04PM (#41928811)
@isodev sorry about that! Just uploaded a video to the README
Re:Show HN: A macOS Client for HuggingFace Chat on Wednesday October 23, 2024 @03:55PM (#41928743)
Yes but you now have the ability to run queries with a shortcut from your Mac, as well as switch to a local model using a dedicated keyboard shortcut. See demo: https://x.com/cyrilzakka/status/1838618605648490974
Re:Show HN: A macOS Client for HuggingFace Chat on Wednesday October 23, 2024 @03:26PM (#41928493)
Should definitely be added to the README but here’s a demo: https://x.com/cyrilzakka/status/1838618605648490974?s=46
Re:Show HN: A macOS Client for HuggingFace Chat on Wednesday October 23, 2024 @03:25PM (#41928486)
Both! You have the option of using HF’s HuggingChat or switching to a local model using a keyboard shortcut
Re:GPT-4 Architecture on Monday July 10, 2023 @10:05PM (#36675623)
not found - is there an archived version I can take a look at?
Re:Med-PaLM 2 reaches a score of 86.5% on MEDQA (USMLE) on Wednesday May 17, 2023 @02:29AM (#35971918)
I see what you mean although I was specifically referring to the delta column. I might be misunderstanding it - can you explain the huge flux in the confidence intervals there? That’s what struck me.
Re:Med-PaLM 2 reaches a score of 86.5% on MEDQA (USMLE) on Wednesday May 17, 2023 @01:39AM (#35971637)
Well not exactly - Table 6 is a bit concerning. It's meant to show that despite significant data contamination between testing and training datasets, the model still performs 'well'. Except look at those confidence intervals - they're all over the place, meaning the model performance isn't very reliable. You have confidence intervals comprising 30-40 percentage points!

Show HN: Phantom Dicom Anonymization

Posted by archiv on Friday January 20, 2023 @10:28AM
from the show-hn dept.

https://github.com/cyrilzakka/Phantom

Hello everyone! I'm a medical doctor working at the intersection of machine learning and medicine.I made a WIP DICOM anonymization tool in Python for my ML workflow. Unlike other anonymization modules, Phantom is whitelist-based, so anything not specified in the configuration file gets automatically purged from the metadata.Phantom also partially supports getting rid of burned annotations in...

( Read More... | no comments | on HN )

Show HN: I made a local transcription app for your Mac

Posted by archiv on Saturday January 07, 2023 @09:37AM
from the show-hn dept.

https://twitter.com/cyrilzakka/status/1611096436661276674

Hello all, I made a small transcription app for your Mac based on OpenAI’s Whisper. Would love some feedback. My plan is to make it easy to load weights from any fine-tuned whisper model to enable specialized dictation for any subfield. It’s still early in development. Thanks!

( Read More... | 2 comments | on HN )