Claims 10B vision model possible in weeks amid Apple hiring
richdotca · x · 2026-08-18
Responding to Apple's recruitment for video reasoning researchers, an insider claimed his team can build custom models with public evals and private data. He noted that GoPro is eyeing their 20MB segmentation models and that they are already running 2B VLMs on edge hardware. He asserted that a 10B multimodal vision model could be built in weeks using 256 competing teams.
More from Models
- Running a fully local AI podcast station with Qwen — sysadmin420 · 2026-08-18
- Built a game in two prompts with Qwen 3.8 — lordekeen · 2026-08-18
- User claims GigaChat 3 Ultra is the best Russian LLM right now — flowersslop · 2026-08-18
- Toast 1 Search Model Released: GPT-5.6 Level Quality at 1/10th Cost — eliebakouch · 2026-08-18
- Qwen 2.5 27B raises questions about scaling laws — Notrx73 · 2026-08-18
- Do smaller LLMs struggle more with multi-step instructions? — orelrevivo · 2026-08-18