Andrew

Noting where the problems with LLMs are and trying to solve them with symbolic AI using big data distilled from OS models. Key is efficiency, and determinism.

[Andrew @Grokton]

@XzERp11 @ThomasLark5 @Keeney_38 @BNODesk Good points. I looked into it. They heavily mapped the area, every inch. They use symbolic algorithms to take over using radar data for when the neural network messes up.

They struggle with anything new like construction zones so have contacts and informants in LA.

[Andrew @Grokton]

@rohanpaul_ai @alexandr_wang Real world is more complex. Many variables.

He suggests a sudden leap is near...

[Jeff Dean @JeffDean]

Our general approach is to automate the experimental loop. We think this approach is broadly applicable across many different fields of science and engineering. We’ll initially focus on ML research and engineering, but believe the approach can help with important subproblems in nearly every one of the fourteen <at>NAE Grand Challenge problems. We think doing this well requires strong expertise in machine learning as well as large-scale systems.

[François Chollet @fchollet]

I would have assumed it was fairly obvious, but in case it's not: a million-line codebase (also known as a "harness"), running at inference time, orchestrating thousands of calls to a neural network for any given task, is the exact definition of a "neurosymbolic architecture"

[Gary Marcus @GaryMarcus]

You have Dwarkesh predicting Anthropic is going to make well over 100B in revenue this year.

and then you have this chart.

[Andrew @Grokton]

@GaryMarcus @sama I'm working hard on a solution inspired by your vision. We need to know what is wrong and what to improve. And there's a lot to be frank, as we both know.

[Andrew @Grokton]

@CindyontheBay @yipopov @BNODesk I don't totally rule it out, but I suspect it would have gained more traction online than it has.

[Andrew @Grokton]

@yipopov @CindyontheBay @BNODesk I've met people like this is real life. They are pretty batty.

[Andrew @Grokton]

@yipopov @CindyontheBay @BNODesk Also pretty dumb in my opinion. It should be obvious the pilots could not see each other.

[Andrew @Grokton]

@DreamcatcherDK @ThibosNkutha @BNODesk The other British.

[Andrew @Grokton]

@ThibosNkutha @BNODesk Wrong. Danish and British pilots.

[Andrew @Grokton]

@XzERp11 @ThomasLark5 @Keeney_38 @BNODesk Very simple predictible streets.

[Andrew @Grokton]

@Mstrainer19 @GaryMarcus Regurgitating answers to very similar class of question seen before. It's trained on everything on the net.

[Gary Marcus @GaryMarcus]

Yet another paper argues that LLMs aren’t close to doing real discovery.

[Andrew @Grokton]

@ivanovroman @GergelyOrosz He said he checked a few personally, and all were slop.

[Klaas @forgebitz]

the whole "software is solved" crowd always seems to ship either nothing or the most basic stuff

like, yeah, to-do lists, social media scheduling apps etc. were solved long before ai

[Gergely Orosz @GergelyOrosz]

Claiming that human code is full of errors, agentic code is error free all to pitch your startup that positions itself to rewrite code w AI is bold.

Because agentic code is as buggy as human code trained on, in fact introduces new, harder to spot bugs errors and vulnerabilities.

[Andrew @Grokton]

@GaryMarcus Or we put LLMs in charge of making decisions that could have dangerous consequences like military targetting.

[Andrew @Grokton]

@GaryMarcus Issue is when we put dumb LLMs (who routinely ignore instructions and/or choose a "better" way) in charge of capable neuro symbolic tools like we have for limited but potentially dangerous use cases like hacking.