Sabitlenmiş Tweet
Rune Berg (bsky: runeberg.social)
905 posts
Rune Berg (bsky: runeberg.social)
@1runeberg
XR specialist and AI tinker. Worked on SteamVR & OpenXR @valvesoftware, @unrealengine, Khronos @OpenXR. Currently exploring intersections between XR & AI
Sydney, New South Wales Katılım Mayıs 2015
114 Takip Edilen563 Takipçiler
@michae1becker Nice! This was one of the things I wanted to do/play with if I had the device. I think utility would be higher if there’s a visual scripting ui so you can easily modify on the fly if it doesn’t have it already.
English
Something *very* cool just launched - ShaderVision, a shader editor for the Apple Vision Pro!
Being able to leverage the actual environment as input geometry is just brilliant. Freemium, but an instapurchase in my book. Gonna have so much fun with this.
English
Open-sourced my standalone interface app for #Ollama & #OpenAI LLMs.
No containers/tools needed. Chats are managed as local json and optionally encrypt with assets.
Executables in releases. No signing certs atm, so consider sponsoring if useful :)
🔗 github.com/1runeberg/conf…
English
Been tinkering w/ a privacy-focused chat app #ConfiChat . Handles local models via @ollama and online @OpenAI w/ local history encryption. Looking for beta testers but planning to open-source this w/e.
Pls let me know if I should keep the "cutesy" branding. Not so sure about it
English
@haagch42 Thanks 👀! Still exploring ollama’s tools as well. It allows easy switching to larger offline and online models and do RAGs - my German class notes might find another use than gathering dust hehehhe
English
@1runeberg Well just larger models will often do the trick already. Just for comparison on e.g. llama3.replicate.dev you can give the 405b model a try for free without an account. They probably slurp up any text you input there, so best to write nothing private of course.
English
@haagch42 Thanks Christoph! I really appreciate that - will split out param 6 then. also sent native friends a longer convo to critique the interaction so I can further tune.
Goal’s to make it “good enough” for personal use, though I wish orgs such as Goethe publishes a “pro” model.
English
@1runeberg @ollama Nice, but the suggestions are kind of nonsensical. I feel like especially smaller models try to hallucinate something that sounds vaguely right if there is nothing else to say. Adding a rule that it shouldn't make stuff up if it has nothing to criticize may or may not help.
English
@pvncher Personally I’d prefer a context sensitive/selectable choice of “AI provider”.
As for using llama, not sure why Apple won’t just develop their own model using exisiting models as benchmarks. They would certainly have the engineering talent to do so (or can build one).
English
@EspireMike I use them, they’re great alternatives to Adobe’s suite.
English
wow, this is awesome 👏
Affinity@Affinity
Try Affinity Photo, Designer and Publisher free for six months – no strings attached! Download the extended free trial from affin.link/6ixx Our CEO Ashley Hewson explains why we’re making this unprecedented offer… #TryAffinity #6monthsfree #timetoswitch
English
@pvncher Agree - a very clear use case of xr that the general public can appreciate.
English
This combined with viewer glasses like the xreal, could start to become a really viable form factor for laptop alternatives
Pirat_Nation 🔴@Pirat_Nation
>AMD Ryzen PC Gets Integrated into a Foldable Keyboard Key features include: •AMD Ryzen 7 8840U APU (8 Zen 4 cores, up to 5.1 GHz, 16 threads, Radeon 780M iGPU) •16 GB / 32 GB RAM •512 GB / 1 TB storage •60 Wh (16000 mAh) battery •Integrated "Mini-Touch" touchpad •Three performance modes: Efficiency, Balanced, and Performance •15 x 10 cm dimensions, weighing 800 grams techpowerup.com/324256/amd-ryz…
English
This got me curious if these models can generate doxygen comments, so I fed #ChatGPT and #ClaudeAI code from my OpenXR wrapper lib and url to the spec.
Claude came up with "ok" results but chatgpt seems unable to access the spec.
Not quite useable but seems getting there imo.
Rune Berg (bsky: runeberg.social)@1runeberg
Been testing @ChatGPTapp & @AnthropicAI side-by-side for some of my more boring tasks. Here's an interesting one using a quick sparse prompt. #chatgpt4o provided an interesting name; #ClaudeAI is I guess an Apple fan 😆 Both hallucinated features but leaning on ChatGPT on this
English
@1runeberg Well, you can make a readme and pin it to your profile.
But I was referring to this x.com/gf_256/status/…
cts🌸@gf_256
so umm... yea lets just say ... github has a css injection 😳
English
@pvncher @ChatGPTapp @AnthropicAI ultimately - i think i'll go with the chatgpt first paragraph one as the initial draft and refine from there. prefer coding than word smithing 😅
English
@pvncher @ChatGPTapp @AnthropicAI This was more a test than an actual prompt - intent was a single paragraph intro for the repo readme.
It'd be interesting though if they can create summaries using the doxygen generated api doc (I doubt it without much direction).
Plus side: some good ideas from the responses.
English
Been testing @ChatGPTapp & @AnthropicAI side-by-side for some of my more boring tasks.
Here's an interesting one using a quick sparse prompt. #chatgpt4o provided an interesting name; #ClaudeAI is I guess an Apple fan 😆
Both hallucinated features but leaning on ChatGPT on this
English
@pvncher Just tried it to create a simple tic tac toe game. Took just four prompts to get it working the way I wanted with sparse prompts. two bugs which it fixed - one was minor UI and another is the "ai" opponent logic.
Not bad at all:
app.screencast.com/O1G0KdWx1dFLa
English
Using the new Claude Sonnet 3.5, it feels like when it writes code, it can visualize what the program’s execution looks like, in a way that other LLMs haven’t before.
I think that’s why it’s impressing coders so much.
English
@pvncher Code check against a repo code style document (e.g. var naming rules, ref for out vars over ptrs, early return on error/checks, etc). AI should be able to do this I reckon much more consistently than a human code reviewer.
English
I’ve been getting into writing GitHub actions that use llms to help you analyze and review code, and I’m thinking of working on something that I’ll open source.
Is this something folks are interested in?
What would your ideal GitHub ai tool do for you?
English
@pvncher ah gotcha. i'm hopeful then - i didnt think my pronunciation was THAT bad 🤣 looking forward for the audio bits to catch up then.
English
Just tried the new ChatGPT model. My German's super rusty so thought I'd see if it can help with practice, from the basics.
Perhaps my pronunciation is super bad as it thought I was speaking Japanese at one point 🤣
Here's the transcript:
chatgpt.com/c/4a06c996-a0f…
English
@pvncher is it not? popped up this morning saying if i wanted to try chatgpt4o, so thought i'd give it a whirl.
Didn't know there was already live audio chat before.
English
@1runeberg The audio parts aren’t live yet, so you’re using the old voice model here.
English