Understanding AI

Waymo has been at the forefront of cutting edge AI/ML since the beginning: https://waymo.com/research/ This is a good read: https://www.understandingai.org/p/elon-musk-wants-to-dominate-robotaxi… This talk by Waymo co-CEO is even better (especially from 33:50): https://www.youtube.com/watch?v=s_wGhKBjH_U

Eh? I’ve never heard of them until this article and my first impression of it was worse than bad. The author seems full of hubris and the subject matter is a reformulated “tech journalism is broken” is comically late by, oh, about a decade. The author seems to be unaware that this isn’t a new thing. Then there’s some cringeworthy stuff like unironically using the word normie before conjuring an irrelevant excuse to proudly remind everyone how easy political writing was in his corner of the society where there is only one correct candidate to support. Author is also selling a paid subscription to an AI magazine. In one of the free articles he coins the phrase “singularism” (and used it in multiple articles) being oblivious to the fact that singularitarianism has existed for 30 years and means the same thing as his “catchphrase”. As the saying goes, “if you can’t pronounce singularitarian, you can’t be one”. https://www.understandingai.org/p/why-im-not-worried-about-ai-taking It’s alright folks, AGI is not a risk because the world isn’t a chessboard. It just seems like a grift festival in there.

> often with no clear winner. You have to pick one somehow. It's not like it's taking some kind of "real" output and obfuscating it. It’s very rare for multiple outputs to actually be equal so the only choice is to choose one at random. Instead its become accepted practice to make sub optimal choices for a few reasons, one of which really is to decrease the likelihood of reproducing existing text. Nobody wants a headline like: “Meta's Llama 3.1 can recall 42 percent of the first Harry Potter book” https://www.understandingai.org/p/metas-llama-31-can-recall-42-percent

They are reducing jobs already. Recent grads are having serious trouble to get work right now: https://www.understandingai.org/p/new-evidence-strongly-suggest-ai

The ones I visit daily https://pluralistic.net/ https://www.wheresyoured.at/author/edward/ https://www.understandingai.org/ https://nlnet.nl/project/current.html https://european-alternatives.eu/ https://www.openrightsgroup.org/ https://hope.net/ for some human spiritual depth: https://www.themarginalian.org/

I checked this out [1] just to get a sense, and apparently there have been 200 total collisions involving Waymos, about one every 100,000 miles. 43% were at relative speeds of under 1 MPH. Only 23 were serious enough to cause an injury or deploy airbags, and of those, 16 were rear-end crashes. At least one of the 16 involved a driver who hit the Waymo a second time as he tried to run away, so probably not good faith. Anyway: might be that Waymos hit the brakes a lot, but does not seem like an epidemic of fast "slam on the brakes" rear-end collisions. [1] https://www.understandingai.org/p/human-drivers-are-to-blame-for-most

Author here. What do you mean? The article this thread is about has hardly any opinion in it. Or consider this piece explaining how LLMs work: https://www.understandingai.org/p/large-language-models-explained-with

> your problem space doesn’t need an LLM to solve it Hmmm… kinda opinion right? I’m saying; in specific situations, you can validate the output and aggregate solutions based on deterministic criteria to mitigate hallucinations. You can use statistical methods (eg. There’s a project out there that generates tests and uses “on average tests pass” as a validation criteria) to reduce the chance of an output hallucination to probability threshold that you’re prepared to accept… for certain types of problems. That the problem space is trivial or not … that’s your opinion, right? It has no bearing on the correctness of what I said. There’s no specific reason to expect that just like you can validate output against a grammar to require output that is structurally correct, you can’t validate output against some logical criteria (eg. unit tests) to require output that is logically correct against the specified criteria. It’s not particularly controversial. Maybe the output isn’t perfectly correct if you don’t have good verification steps for your task, maybe the effort required to build those validators is high, I’m just saying: it is possible. I expect we’ll see more of this; for example, this article about decision trees —> https://www.understandingai.org/p/how-to-think-about-the-openai-q-rumo… requires no specific change in the architecture. It’s just using validators or search the solution space.