humanShaped

[humanShaped @_humanShaped]

@UnitreeRobotics It's an expensive piece of junk. Just pre-programmed routines + balance with no awareness of environment or adaptivity. Nobody has any use for this crap.

[humanShaped @_humanShaped]

@47fucb4r8c69323 It's hard to be dangerous when you don't exist.

[humanShaped @_humanShaped]

@kimmonismus This is part of their data collection project where they're collecting data for training from teleoperation. All of that recent hospital footage was too. It doesn't have the autonomy for this (and certainly didn't when this was recorded).

[humanShaped @_humanShaped]

@kimmonismus It's teleoperated and plugged into a short cord. Not autonomous, not really useful.

[humanShaped @_humanShaped]

@Scobleizer @NuroAi_Official This whole discussion is how a tech guys loses credibility with anyone reasonable.

[humanShaped @_humanShaped]

@carl_feynman @BellikOzan I just now conceived of a bigger one; 101 million kilometers long. I don't think anyone could conceive of a bigger one now.

[humanShaped @_humanShaped]

@liron I'm not claiming this is an answer, and you probably know this already, but he said it was a general computation device. The inputs and outputs are anything you like to run an arbitrary program. But the programming paradigm is radically different in return for reduced heat.

[humanShaped @_humanShaped]

@ZajcGal @MatthewBerman Just a guess, but they could mean that errors were reduced to 1/7th of before, but it doesn't sound as good. A lot of people do this. No comment on whether this is a good idea.

[humanShaped @_humanShaped]

@gaurav_dhiman @MatthewBerman 4x is a 300% increase. It is 400% of its previous speed, but a 300% increase.

[humanShaped @_humanShaped]

@liquidai No public weights, but it's just mediocre performance. So it will probably go nowhere. What a waste of time.

[humanShaped @_humanShaped]

@yumidiot Maybe they will drop Blueberry, the text-to-image model (or expand the multi-modality of gpt-4o or o1-*).

[humanShaped @_humanShaped]

I'm building a better front-end app for AI LLMs.

It's cross platform, native and supports @OpenAI, @Anthropic, @AIatMeta, @Google_Gemini_, @GroqInc, @MistralAI, @ollama, llama.cpp, etc.

Characters, tools, multi-bot chats, agents, low latency speech...

https://t.co/vbezr0P9XO

[humanShaped @_humanShaped]

@ryanaraine @SecurityWeek Sounded like an unhandled exception crashing it rather than a buffer overflow or similar, so this is probably untrue.

[humanShaped @_humanShaped]

@TechCrunch A non-functional mockup is not a reveal any more than a cardboard cutout is. He's just pulling his Theranos style lie-until-you-make-it stunt yet again.

[humanShaped @_humanShaped]

@MatthewBerman Yes, they can join the Theranos and Tesla club of lies, essentially.

[humanShaped @_humanShaped]

@apples_jimmy @tszzl @OfficialLoganK Okay, apparently not. Sam jumped the gun this time.

[humanShaped @_humanShaped]

@apples_jimmy @tszzl @OfficialLoganK Maybe July 23 they will release an open weights tiny model for mobile devices, or something else non-GPT-5-ish, mainly to rain on the Llama-3-405B parade. Sam loves rain and parades.

[humanShaped @_humanShaped]

@BenjaminKlieger @GroqInc And it so it works with Crosstalk AI, but Groq do have a few custom changes too.