tl;dr Waymo robotaxis will stop when they see a person wearing a t-shirt with a stop sign on it. Surprising?

What will they do if they see a person wearing a 'Speed Limit 60' t-shirt in the middle of a San Francisco street?

thestreet.com/automotive/the…

This talk from Gary Marcus is quite compelling (and pleasingly accessible) about the limitations of current generative AI. He suggests that greater scale of current tech won't get us much further. youtube.com/watch?v=zj0Ek3…

The intuition behind the “scale is all you need” movement was that it would be trivial to build AGI if you just kept gathering more data and building bigger models.

What we are actually seeing is that it is apparently hard to get far beyond GPT-4, even with immense $ invested.

We need a Consumer Reports or Underwriters Lab for AI testing. All the public benchmarks are game-able and mostly not useful measures of things LLMs do. We need secret test batteries for subject areas (coding, reasoning, human conversation, writing) and secret red team tests, too

Agreed. Level 5 autonomy is the only way to go. Will need expensive LiDAR etc for now. Can’t expect humans to suddenly intervene if the level 3/4 system messes up. This cavalier approach from Tesla may kill many innocent bystanders, and may trigger legislation which could kill

⚠️ Perhaps the biggest risk facing the AV industry is that people will be led to trust vehicles that are not fully ready for prime time.

Here, a Tesla driver appears to have put to much faith in FSD, and to have killed a motorcyclist in the process. The Tesla driver has been

Cognitive psychology tidbit of the day (see my reply for the answer).

Amazing that a chip bricked in 46-yr-old Voyager I, preventing it from sending data, and NASA figured out how to split up and reallocate its functions to other hardware, sending code 15 billion miles away (45 hours round trip!)--and Voyager's back online. blogs.nasa.gov/voyager/2024/0…

Really great engineers don’t make wasteful, underwhelming products like generative AI; they design chips and systems that work in space for 46 years before failing … and can still be fixed when they are 15 BILLION miles away.

Color opponency: the can appears reddish even though it’s only black and white dots surrounded by cyan.

Tesla driver arrested for Vehicular Homicide after striking and killing a motorcyclist while on auto pilot. kiro7.com/news/local/cha…

People who watch *the exact same* video of a protest often come to radically different conclusions.

People's views of protestors and the police depend on the congruence of the protestors’ positions with the subjects’ own worldview jstor.org/stable/41511108

A foolish failure to delay gratification in order to rush premature technologies can cross party lines, claims Beff Jezos — e/acc ⏩. More or less.

There is something emblematic about adopting another person’s entire wardrobe and even mannerisms in the age of LLMs.

Not wrong.

Will it count as “just desserts” when the people now claiming there are no downsides to LLMs discover in 2026 that their models no longer work well because of all the crap their models spewed and subsequently ingested in 2023-2025?

more AI-induced internet #enshittification
cc Cory Doctorow NONCONSENSUAL BLUE TICK

Is it just me, but when, in the context of ChatGPT someone asks what the GPT stands for, you are tempted to say Generalised Plagiarism and Theft?