Links for 2022-12-04
Talking to Robots in Real Time: “Interactive Language shows initial evidence that large scale imitation learning can indeed produce real time interactable robots that follow freeform end user commands.” https://ai.googleblog.com/2022/12/talking-to-robots-in-real-time.html?m=1
FDA gives first safety sign-off for company selling lab-grown meat https://www.nbcnews.com/science/science-news/fda-gives-first-safety-sign-company-selling-lab-grown-meat-rcna57536
“Things are going to get weird. Got a custom AI video avatar of myself from @synthesiaIO based off of 2 minutes of video material 🤯” https://twitter.com/siavashg/status/1597588865665363969
Compressing Volumetric Radiance Fields to 1 MB: “…our method can achieve a compression ratio of 100× by reducing the overall model size to 1 MB with negligible loss on visual quality.” https://arxiv.org/abs/2211.16386
TSMC will invest around $32 billion in a 1nm-capable fab. That's up from around $20 billion for N5 and N3 fabs that the company currently operates. TSMC has outlined plans to start making chips using its N2 technology in the second half of 2025. https://www.tomshardware.com/news/govt-official-tsmc-looking-at-dollar32-billion-investment-for-1nm-fab
Google’s Sycamore chip: no wormholes, no superfast classical simulation either https://scottaaronson.blog/?p=6871
Like humans, macaque monkeys have brain areas devoted to processing emotional body language. These areas respond more strongly to body postures indicating fear than to neutral body postures – not only in other macaques but also in humans and cats. https://science.org/doi/10.1126/sciadv.add6865
Genetically modified tobacco plant produces cocaine in its leaves https://www.gwern.net/docs/biology/2022-wang.pdf
Could a single alien message destroy us? https://youtu.be/st9EJg_t6yc
"How persuasive is AI-generated argumentation? An analysis of the quality of an argumentative text produced by the GPT-3 AI text generator", Hinton & Wagemans 2022 https://content.iospress.com/articles/argument-and-computation/aac210026
China operating over 100 police stations across the world with the help of some host nations, report claims https://edition.cnn.com/2022/12/04/world/china-overseas-police-stations-intl-cmd/index.html
“Harvard employs 7,024 total full-time administrators, only slightly fewer than the undergraduate population. What do they all do?” https://www.thecrimson.com/article/2022/11/29/anderson-bureaucratic-bloat-harvard/
I've now seen a bunch of tweets along the lines of “ChatGPT could pass my class”. In light of expected progress, and even after discounting hyperbole, this should really make people think long and hard about the future.
For example, is it really worthwhile studying professions that can be mastered by ChatGPT? And even if your profession is safe for now, how confident are you that the next AI winter will set in before it gets smart enough to master it as well?
Maybe people like Gary Marcus and Judea Pearl are right in claiming that the current machine learning techniques are dead ends, and progress will soon hit a wall. These are the people telling us that the metaphorical comet will miss Earth. But I can't help it, and look up in eerie anticipation.