leumon

7 ms·

121.
by leumon 3y ago
"When in doubt, scale it up." - openai.com/careers
122.
by leumon 3y ago
I wonder how gemini 1.5 compares to its open source variant (that also has video input) released about the same time as gemini: https://largeworldmodel.github.io/
123.
by leumon 3y ago
For writing text (emails, etc.) and asking basic questions gemini is superior to gpt-4 in my opinion. But when it comes to logic, programming and resoning, gpt4 is better in my opinion. It rarely makes syntax or logical errors, and the jupy
124.
by leumon 3y ago
If you have a decent gpu (16gb+ vram) and are using Linux, then this tool I wrote some days ago might do the trick. (at least for googles recaptcha). Also, for now, you have to call the main.py every time you see a captcha on a site and you
125.
Show HN: Recaptcha Solver using LLaVA-v1.6(github.com)
3 points by leumon 3y ago|0 comments
126.
by leumon 3y ago
actually from a prompt engineering perspective it makes sense to use words like "please" because the llm is more likely to give better results (which it learnt from the texts it was trained on)
127.
by leumon 3y ago
Would be cool if we sometime would get https://github.com/seemoo-lab/opendrop as an android app (but it would only work with rooted devices). This would allow to send and recieve files to iPhones from android via Apple
128.
by leumon 3y ago
I actually did a bit of research and found this prompt: Please translate the user message from {src} to {tgt}. Make the translation sound as natural as possible. This actually gives much better results, and now I think I actually changed my
129.
by leumon 3y ago
I still think (from limited testing with german->french and vice-versa) that DeepL gives better results than gpt4.
130.
by leumon 3y ago
I agree but I think it's a problem that they are using it wirelessly. I hope they have backup batteries..
131.
Missing Titanic submersible built with NASA help(space.com)
1 points by leumon 3y ago|0 comments