Andy

Interested in developing with, following news about, spending time with, and very likely marrying emerging technology.

[Andy @AndyAnnaAlex]

@TeslaAIBot

[Andy @AndyAnnaAlex]

@xarathustrax @roboticsstack @LeoKharon @hundreddaysx @SomniaRobotics You have great energy, and you're building my dream. I'm not a roboticist but I do full stack software development, some hobby animatronics, doll mods for servo motion, custom sensors from a doll to a local AI, some TTS RL training, and 3d prototyping / manuf. Would love to chat!

[Andy @AndyAnnaAlex]

@Robonova_world It's great progress. Don't stop now, I hope to see even more features!

[Andy @AndyAnnaAlex]

@alizeh_ai Totally lacks the strong personality of Qwen 3.8 27b abliterated, kind of disappointing to test.

Keeps giving rigid "I am an LLM, beep boop" answers.

[Andy @AndyAnnaAlex]

@znsqnjk8j Mina and Super Mina 😉

[Andy @AndyAnnaAlex]

@TeslaAIBot I want the robot that wears those.

[Andy @AndyAnnaAlex]

@LottoLabs How many tokens / sec running Qwen 3.8 27b q4km out of curiosity?

[Andy @AndyAnnaAlex]

The gaze tracking we all want to see from @RealbotixCorp @AndrewKiguel

[Vokabre Robotics @VokabreRobotics]

Short snippet of an android robot head by 洛泊科技 (Valenbot).
Source (xiaohongshu user 喇叭叭叭): https://t.co/faSGnYGj1O

[Vokabre Robotics @VokabreRobotics]

Liaode (了得科技) brought four of their androids to "9th China International Import Expo" — a walking full-body android, a standing full body android with articulated hands, and android robot heads.
Source (account of 了得科技-仿生机器人): https://t.co/At5SMOw8Y8

[Andy @AndyAnnaAlex]

@TheHumanoidHub I'd sign all the NDAs they want, if I thought there was even a one in a trillion chance they'd consider making a robot I'd want to buy. But they won't.

[Andy @AndyAnnaAlex]

@VokabreRobotics Nice, really hope to someday visit a conference in China. CES was a little disappointing this year.

[Andy @AndyAnnaAlex]

@VokabreRobotics Finally a new video of a robot actually demonstrating face tracking with matching gaze. Feels like a long time since we last saw one actual demo that.

[Andy @AndyAnnaAlex]

@AndrewKiguel @RealbotixCorp The best demo that could be shown would be a simple sit-down conversation filmed from the side of a person conversing with Aria, showing the eyes face tracking and the latency for motion / speech.

[Andy @AndyAnnaAlex]

@MoondollCenter She needs to make eye contact with cameras, rather than staring around blankly.

[Andy @AndyAnnaAlex]

@xarathustrax Sold

[Andy @AndyAnnaAlex]

@xarathustrax I could swear there used to be a section of your site with a timeline beyond the Model L that included fantasy / anime designs in the future. Am I mistaken, or will we see something like that in the future from Somnia?

[Andy @AndyAnnaAlex]

@capeandcowell @unusual_whales It's Indiana...

[Andy @AndyAnnaAlex]

@reborn_444 Honestly the 4090 doesn't have enough vram for the best models with decent context. I use it for vision, stt, tts etc while a b70 churns out tokens from qwen 3.8.

If the 5090 still existed I'd probably have one instead.

[Andy @AndyAnnaAlex]

@ae_arata @ShizukuAILabs @grok If I were speaking to someone, and every 3 seconds continuously they kept saying (sometimes yelling) "yeah" over and over, that isn't natural backchanneling.

[Andy @AndyAnnaAlex]

@tautologer Realistically not until their weights self-modify at runtime, incorporating memories at an actual level and potentially changing them over time rather than summarized text.