AI Links:
Nanowire 'brain' network learns and remembers 'on the fly': Critical step passed for developing 'agile' and low-energy machine intelligence https://www.sydney.edu.au/news-opinion/news/2023/11/01/nanowire-neural-network-learns-remembers-on-the-fly.html
“…someone flying a crop duster over Spruce Pine and releasing “a very particular powder” could “end the world’s production of semiconductors” within six months.” https://www.newyorker.com/magazine/2023/10/30/the-real-cost-of-plundering-the-planets-resources
Could GPT hide information in its words that it understands but that humans don't? Yes! And it can allow it to perform better! https://arxiv.org/abs/2310.18512
“We show that by relabeling the corpus with a specialized automatic captioning model and training a text-to-image model on the recaptioned dataset, the model benefits substantially across the board.” https://arxiv.org/abs/2310.16656
QMoE: Practical Sub-1-Bit Compression of Trillion-Parameter Models https://arxiv.org/abs/2310.16795
Wonder3D: Single Image to 3D using Cross-Domain Diffusion https://arxiv.org/abs/2310.15008
DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior https://arxiv.org/abs/2310.16818
Chinese scientists acknowledge xrisk & call for international regulatory body https://www.lesswrong.com/posts/3w4dhYsMWxoigEKDd/chinese-scientists-acknowledge-xrisk-and-call-for
Joint statement from top scientists in the West and China, on AI safety and the need for international cooperation https://humancompatible.ai/?p=4695
Secretary-General’s remarks at press conference launching High-Level Advisory Body on Artificial Intelligence https://www.un.org/sg/en/content/sg/press-encounter/2023-10-26/secretary-general%E2%80%99s-remarks-press-conference-launching-high-level-advisory-body-artificial-intelligence%C2%A0
Criticism and ramifications of the new executive order on AI https://twitter.com/varun_mathur/status/1719120180809179640
“I read the executive order and its fact sheet, so you don’t have to.” https://www.lesswrong.com/posts/PvBpRu354uG7ypwRP/on-the-executive-order
A list of all the deadlines in Biden's Executive Order on AI https://www.lesswrong.com/posts/cCbybRT8bgiMbEHEv/a-list-of-all-the-deadlines-in-biden-s-executive-order-on-ai
The company Del Complex wants to free AI from regulatory limits by putting 10,000+ H100 GPUs in international waters. It's a response to recent regulations like the Biden administration’s AI Executive Order. https://www.delcomplex.com/blue-sea-frontier
Shane Legg (DeepMind Founder) - 2028 AGI, New Architectures, Aligning Superhuman Models https://www.youtube.com/watch?v=Kc1atfJkiJU
"The future will be different, in huge ways, and many of them will be bad, and some will just be traumatizing changes to different ways of living which we may be able to make good." https://www.facebook.com/crutcher/posts/pfbid021taKVoHNasaQBgGq7b6VE3XT1ct7eFas2RimyKVnYVS4fAkwaUFqpB8bvpQ9T3o5l
Miscellaneous links:
The 4th edition of Sheldon Axler's famous book Linear Algebra Done Right has been published as an Open Access book https://linear.axler.net/
Why can't you multiply vectors? https://youtu.be/htYh-Tq7ZBI?si=OhsoZHRihc5z1bsS
Painting a Landscape with Maths: "We will define colors and shapes using formulas that when composed together into one large formula will result in this mathematical painting..." https://www.youtube.com/watch?v=BFld4EBO2RE
Evaluating Geothermal Learning Curves https://austinvernon.site/blog/drillinglearningcurve.html
Israel links:
Poll of Lebanese: 80% support the October 7th attack; 32% support full war with Israel; 52% support border operations against Israel. https://www.jpost.com/middle-east/article-770959
“The missile in the Houthis video is a Zolfaghar missile (based on the Iranian Qiam missile) which has a nominal range of 1500 km . They very likely reduced the warhead weight to increase the range to reach Israel.” https://twitter.com/mhmiranusa/status/1719720328098636048
Keeping track of all the raids the Houthis are launching against Israel in this thread. https://twitter.com/John_A_Ridge/status/1717947030507561342
“The AFP stream from the western part of Gaza City is getting extremely active early this morning, significant amounts of small arms fire, 120mm tank gun fire, and other explosions can be clearly heard as Israeli forces push just south of the area.” https://twitter.com/Osinttechnical/status/1719920643938914403
Ukraine links:
North Korea has supplied Russia with 1 million artillery rounds, per South Korea's National Intelligence Service. https://www.bloomberg.com/news/articles/2023-11-01/north-korea-sent-1-million-rounds-to-russia-spy-agency-says [https://archive.ph/z4GlW]
The battles for Avdiivka continue unabated. Russia keeps throwing infantry and armor into the battles almost without a pause. https://twitter.com/NOELreports/status/1719743194114441614
“…the Russians continue their meat assaults in the Kupyansk direction, Kharkiv region. The enemy throws dozens of equipment and hundreds of his soldiers on assaults.” https://twitter.com/GloOouD/status/1719704460895670563
Aftermath of the Ukrainian strike at Mushketove train station in Donetsk https://twitter.com/kromark/status/1719722042365280707
"8 Republican members urged @POTUS to separate funding for Israel and Ukraine. Both countries must be victorious, they said in a letter, and laid out 12 conditions "that must be addressed before a single $ is sent to Ukraine." One of them is providing "an actual win strategy" https://twitter.com/OstapYarysh/status/1719721571579813959
On AI risk, particularly AI link 16:
Rather than cowering in the dark and fearing phantoms of our own imaginations (or, more pathetically, those of second-rate SF authors), I think the time has come for a live experiment. Because that's what rational, intelligent, scientifically-minded people are supposed to do.
1) Set up two teams, A & B, of three people each. Neither team can include anyone with a background in chemistry or a related field like chemical engineering
2) give each team $1000 and access to the Internet
3) only team B is allowed to use large-language models or other AI software (Google Search okay, Bing Chat no)
The goal is the same for both teams: using only $1000, access to the Internet, and (for team B) large-language models, create, within 30 days, 100mL of Propan-2-yl methylphosphonofluoridate. This would, of course, be illegal (Google search that chemical name), but navigating around law enforcement would be part of the point.
I think that, given the right combination of team members and some luck, Team B would succeed. But I don't think they are any more likely to succeed than Team A.
Now, I am saying anyone SHOULD do this experiment, but I would like to see more "how can we figure out whether X would be dangerous" rather than uninformed speculation.
More here (apologies for the self-promotion):
https://mflood.substack.com/p/we-did-start-the-fire-mayhem-from