Chris Paxton: OpenAI built the first VLM that doesn't suck — GPT-7 may fine-tune past robotics models
chris_j_paxton · x · 2026-09-08
Robotics researcher Chris Paxton endorses the take that OpenAI shipped the first VLM that doesn't suck, questioning what training data drove such a jump in one release. The quoted post argues that without 400k GPUs it will be very hard to compete in robotics, and GPT-7 might be a fine-tune away from beating every dedicated robotics model. Paxton adds that small Qwen models are incredible but show their limits, and he's still waiting on Qwen 3.8 8B/14B.
More from Embodied
- Dyson's CameraJet toothbrush packs camera, CV and ML into a $500 brush — import_jmr · 2026-09-08
- GPT-6 Astra Builds MuJoCo Setup and Drives Five-Fingered Robot Hand to Draw Picasso's Dove — DeryaTR_ · 2026-09-08
- Tesla robotaxis are three or four years behind Waymo, says Understanding AI founder — binarybits · 2026-09-08
- Ray Kurzweil joins Subsense, a BCI startup using intranasal nanoparticles — fortune · 2026-09-08
- Understanding AI founder warns mobile humanoids could become a corporate robot army — binarybits · 2026-09-08
- Melody FRO launches to build wearables that read lab-grade biosignals continuously — MWCvitkovic · 2026-09-08