Christian Kothe @christiankothe@social.cleverlibre.org

Reflecting on the chunky cables featuring in Half-Life 2 and later, I came to the conclusion that Combine electrical systems must clearly be running on 1.5 Volts!

@robpike When I got to "One standardized version of G-code, known as BCL (Binary Cutter Language), is used only on very few machines." I could suddenly sense the desperation Owen Lars must have felt as he found himself without a droid that speaks the binary language of moisture vaporators.

@grumpygamer Know the "bus factor"? Maybe we should call it the 737 factor now

... the wild and probably bogus details aside though, I've never bought into the idea that hallucinating or BSing is an unsolvable intrinsic flaw of LLMs, since it may take not much more than operationalizing the process we humans use to construct an internally consistent world model, which is to explore a range of consequences that follow from our beliefs, spot inconsistencies, and update our world model accordingly. And that looks like something that could be attempted in well-trodden paradigms like RL or GANs or something that's not much more complex, so my bet would be that we should've largely worked it out within 4-5y.

Woke up from a strange vivid dream this morning in which I was attending an ML symposium and someone gave a talk on overcoming the hallucination problem with LLMs. The main slide had a graph of nodes representing LLM statements and they were doing some sort of graph diffusion process where the "curl" operator was pinpointing the contradictory/inconsistent statements, which they could then follow to update the weights to discourage those from occurring. Needless to say I immediately tried to arrange an improptu mtg between the speaker and some DL luminaire who was also there to get them to adopt it.😂

@simon Maybe the default prompt of these things should be written by a motivational speaker. Like, "You're extremely focused, you're about to do the work of your life, ..."

@BartWronski Yeah it's an interesting observation. There's a utility factor with mobile that's just out of reach for the mainly experience-oriented nature of VR. I was one of those 2011 VR nerds, and I was one of the 2007 AI nerds, the 2011 ray tracing nerds, etc. There's probably going to be a time (my nerd-self will say likely in our lifetimes) when you can get some neural implants and join some matrix-like social network, and that'll probably check the box of "I NEED that" for a lot of folks seeking those connections. It's clearly not the pair of googles and the motion controller though, and of course a scenario that seemed appealing 20 years ago feels increasingly dystopian these days.

@BartWronski Not sure about the next big thing, and the lack of a definitive killer app is perhaps a testament to the power of good old-fashioned screens to immerse our minds. That said, HL:Alyx was for the most part as immersive as it gets, and I've probably spend 100+ hrs randomly exploring in Google Earth VR, hours at a time - I couldn't imagine doing anything close to that with Google maps (or earth). Those 2 experiences made it worth the expense for me, and things like Eleven Table Tennis VR take it further yet (yes, one could go find a club for that IRL etc etc, but that's a lot of overhead in comparison).

If you haven't come across this nice science-related article: https://arstechnica.com/science/2024/01/the-key-to-fighting-pseudoscience-isnt-mockery-its-empathy/

Wondering if anyone out there is using LLMs are a proposal heuristic in NAS. Would seem fruitful (e.g. after fine-tuning on NeurIPS). Add in reinforcement learning for bonus points. It's not quite recursive self-improvement since re-architecting and retraining the LLM would be a slow, expensive, and human-in-the-loop step.