symisc_devel

I do believe this will be the norm from now on to get access to top frontier model. Computing capacity plus state restrictions plus KYC will be imposed to organisations to get access, individuals will be served last on the queue with degraded performance. Once the Chinese models catch up, nobody (at least individuals) will turn back again to frontier labs.

BMW's latest infotainment despite being intimidating for first time users is quite decent and intuitive compared to the horrors I saw from other German car makers.

We got a $12,000 bill a few hours ago on a presumably leaked gemini API which I very much doubt, and we are trying to resolve the issue with a real support agent. I think they messed something internally and customers are getting these bills.

It is 100% vibe coded.

5.3 codex is a quite good coding agent for complex tasks.

Gemini 3 pro is way better than Opus especially for large codebases.

There is already a C library that does realtime ascii rendering using décision trees:

GitHub: https://github.com/symisc/ascii_art/blob/master/README.md Docs: https://pixlab.io/art

Congratulations for the launch. Actually we launched a similar product recently named Vision Workspace (https://vision.pixlab.io). The general chat niche is quite saturated and practically locked by the major players. I recommend that you focus on one core feature and pivot from there. For us it was the built-in OCR and document query interface inside the UI that initiated the traction and the app is quite popular now in Japan and Malaysia.

I work on the PixLab prompt based photo editor (https://editor.pixlab.io), and it follows exactly what you type with explicit CAPS.

Imagine paying $200/mo for this privacy nightmare

This issue is notorious for BMW cars. You have to notify the ECU each time you install a new battery.

Well, they are relatively easy to spot with the current AI software used to generate them especially if you are dealing on a daily basis with presentation attacks aka deepfakes for facial recognition. FACEIO has already deployed a very powerful model to deter such attacks for the purpose of facial authentication: https://faceio.net/security-best-practice#faceSpoof

Open source GUI libraries are lacking behind the gate locked, closed ones like Adobe. Even Macromedia UI back in the days 20 years ago looks way more appealing and polished than the current open source offering. The only polished open source UI in my opinion is Blender but apparently they have their own rendering engine built from scratch just like Adobe.

Shameless plug: We released an embedded version of SD named Tiny Dream: https://github.com/symisc/tiny-dream which let you generate high definition output images (2048x2048) in less than 10 seconds, and consumes less than 5GB per inference unlike this one which takes 11 hours to generate a 512x512 pixels output despite being memory efficient.

PixLab (https://pixlab.io) & FACEIO (https://faceio.net) | Full-or-part-time | Remote | Computer Vision / Full stack Engineers |

PixLab, a leading provider of Machine Vision, Face Recognition & Media Processing APIs is looking for:

* Embedded C & Computer Vision engineer(s) to work on the SOD (https://sod.pixlab.io), embedded computer vision library.

* Senior Python engineer with proficiency in PyTorch to work on FACEIO (https://faceio.net), our facial authentication web framework for web sites & apps.

* C++ developer with ML expertise to work on the port of Tiny-Dream (https://pixlab.io/tiny-dream), our embedded Stable Diffusion C++ library from ncnn to ggml.

* React/Vue JS Web developer(s) with expertise in fabric.js to work on a brand new, web based photo editing software backed by generative AI.

Reach out to Vincent via contact AT pixlab.io with your resume if interested.

Hi HN,

Tiny-dream is designed to be embedded on larger codebases (host programs) with an easy to use C++ API.

The project github is located at: https://github.com/symisc/tiny-dream The C++ API documentation is located at: https://pixlab.io/tiny-dream#cpp-api

The current tensor engine is backed by ncnn with planned transition to ggml in the short period. Nevertheless, in our experimental ggml port, we found out that ggml doesn't quantize well, and is better suited for LLMs than heavy computer vision tasks. You can refer to the roadmap page at https://pixlab.io/tiny-dream#roadmap for more information.