undefined | Better HN

0 pointsjiggawatts1y ago0 comments

They may not need direct subscription revenue to recoup their investment.

A variant of multi-modal LLMs may be the solution to self-driving cars, home robotics, and more.

I keep saying that to be a really effective driver, an AI model will need a theory of mind, which the larger LLMs appear to have. Similarly, any such model will need to be able to do OCR and read arbitrary street signs, and understand what the sign meant. Most modern LLMs can already do this.

0 comments

steve_adams_861y ago

Since when do LLMS appear to possess theory of mind? The papers I've read on this show impressive capabilities, but only within conditions where language-based training would strongly indicate appropriate outputs for the LLM to use. Outside of familiar conditions, or even realistic conditions, they don't do well. It's all still statistical string-building.

sebzim45001y ago

Can you give an example prompt that shows a lack of theory of mind? It used to be easy, but now that models have improved I've noticed that the 'LLM doubters' nowadays are forced to make generic, almost spiritual, claims rather than making concrete statements.

emptysongglass1y ago

The burden isn't on us, it's on people making claims that LLMs do possess theory of mind. You're the one who has to prove it, not us.

3 more replies

HarHarVeryFunny1y ago

You're not going to run a SOTA LLM of this size off batteries (robotics), even in a car where the alternator is charging them, nor can you afford to rely on a high-speed internet connection being available 100% in a life or death (FSD) application.

I don't have so much faith in the future of current-architecure LLMs, but I do think that AGI will be needed for safe FSD and general-purpose robots that need to learn and operate in an uncontrolled environment such as a home.

ggreer1y ago

A typical car alternator outputs 1.5-3kW of electricity, and EVs can output arbitrary amounts of power for electronics (though that will reduce range). That's more than enough to run purpose-built circuitry for a SOTA LLM. For a home robot, you could run the compute in the home instead of in the robot's body.

I don't think AGI is needed for FSD because we already have safe FSD in the form of Waymo, and competitors aren't far behind. People forget that self-driving doesn't have to be perfect. It just has to be better than human drivers. Human drivers get sleepy, drunk, angry, and/or distracted. They can't see in 360º or react in milliseconds. Most cyclists and pedestrians prefer current FSD implementations to human drivers, as the computer-driven cars are much better about yielding, giving a wide berth, and being patient.

HarHarVeryFunny1y ago

Waymo is obviously pretty decent, but it's easy to drive 99.9% of the time. It's when there's invisible black ice on the road, or an animal runs out in front of you, or you lose visibility due to sun glare or whatever (I once had windshield wiper fluid = water flash freeze on contact) - maybe mud on a camera, or a wheel falls off your car or the one in front, etc, etc, that things get weird.

Having autonomous cars that are statistically safer then the average driver is a reasonable bar to allow them on the road, but for ME to want to drive one I want it to be safer than me, and I am not a hot-headed teenager, or gaga 80-yr old, or drunken fool, and since I have AGI (Actual General Intelligence) I react pretty well to weird shit.

j / k navigate · click thread line to collapse

0 comments

steve_adams_861y ago

sebzim45001y ago

emptysongglass1y ago

The burden isn't on us, it's on people making claims that LLMs do possess theory of mind. You're the one who has to prove it, not us.

3 more replies

HarHarVeryFunny1y ago

ggreer1y ago

HarHarVeryFunny1y ago

j / k navigate · click thread line to collapse