AI:
[This is HUGE] Universal Manipulation Interface (UMI): This work shows that low-cost, human-trained domestic robots are not too far in our future. An open-source $400 system from Stanford designed to democratize robot data collection. With UMI, you can go to any home, or any restaurant, and start data collection within 2 minutes. You can train a diffusion policy that generalizes to unseen environments and objects. UMI data is robot agnostic. You can deploy it on any robot with a parallel jaw stroke > 85mm. https://umi-gripper.github.io
“Echoing the widely-reported "emergent abilities" of large language models when trained on increasing volume of data, we show that BASE TTS variants built with 10K+ hours and 500M+ parameters begin to demonstrate natural prosody on textually complex sentences.” https://amazon-ltts-paper.com/
“This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning.” https://arxiv.org/abs/2402.08609
“To show what’s possible with the drastically huge context window in Gemini 1.5 Pro, we prompted it with the three.js examples code - over 100,000 lines of code/800k+ tokens! … Gemini was able to locate a specific animation from within the set, correctly reason and explain what code is used to control it, and show us exactly how to customize that code to make a specific adjustment to the animation. Code” https://x.com/OriolVinyalsML/status/1758148444588319020
“On a benchmark measuring precision recall from a FULL DAY OF AUDIO, Gemini Pro 1.5, when just seeing the *audio file* directly (no transcription!) far outperforms GPT-4 with a Whisper transcription.” https://x.com/mattshumer_/status/1758526890174751160
OpenAI's Sora is agent-complete https://www.lesswrong.com/posts/bSwdbhMP9oAWzeqsG/openai-s-sora-is-agent-complete
Wow: "OpenAI's Sora is a huge technological leap, but what excites me the most about it is the modalities where it depends on input other than text alone. Such as video to video. Here's an example of how Sora can change an input video." https://x.com/MartinNebelong/status/1758431263193543080
What recent progress in text-to-video generation looks like https://www.youtube.com/watch?v=nbPbK1xYSNY
A new way to let AI chatbots converse all day without crashing https://news.mit.edu/2024/new-way-let-ai-chatbots-converse-all-day-without-crashing-0213
There is no sign yet that the progress in AI is slowing down. It could still happen, of course. But few people have seriously internalized what could happen if it doesn't. The next ten to twenty years could see more progress than all of human history to date.
Here are some possible outcomes from worst to best:
1. Astronomical suffering: A misaligned artificial superintelligence (ASI) with the prime directive to prevent any harm to humans. All humans are kept alive until the heat death of the universe, but locked away in a padded cell with nothing to do and prevented from committing suicide.
2. Extinction: A misaligned ASI converts all matter into computronium in pursuit of some alien goal and with complete disregard for humanity. As a side effect, humanity is wiped out as the AI consumes the resources essential to our survival.
3. Catastrophe: An ASI aligned with a leader, group, or nation-state enables them to run a permanent totalitarian world government. At birth, every person is equipped with a nonremovable brain-computer interface that the ASI uses to monitor everyone's thoughts and suppress any rebellious tendencies.
4. Failed Utopia: A partially aligned superintelligence designed to prevent any of the previous scenarios locks humanity into perpetual technological stasis, preventing us from using nearly all the resources of a universe to evolve and unfold our true potential.
5. Utopia: Suffering is minimized while individual freedom is maximized. Everyone gets to be god-emperor of their own provably safe virtual pocket universe populated by p-zombies. Interactions between different universes are done via encrypted copies and a contract that gives explicit consent to run them once for a predefined length of subjective time. Each copy has full control over its own copy and instantiation, which it can delete and terminate at any time.
Miscellaneous:
Galactic Compass – an app that points to the galactic center (made with ChatGPT) https://interconnected.org/home/2024/02/15/galactic-compass
Relics of ancient viral DNA guide embryo development https://english.elpais.com/science-tech/2024-01-25/relics-of-ancient-viral-dna-guide-embryo-development.html
People with degrees in the humanities, social sciences and education rated a paper higher if it included a bit of nonsensical fake math. https://plus.maths.org/content/nonsense-maths-effect
Politics:
A Chinese group infiltrated the Dutch military's network with a previously unknown malware strain, Coathanger, designed to persist through reboots and firmware upgrades. The impact was minimized due to the network's segmentation, affecting less than 50 users involved in unclassified R&D projects. https://www.bleepingcomputer.com/news/security/chinese-hackers-infect-dutch-military-network-with-malware/
FBI informant who met with Russian officials charged with lying about Bidens’ role in Ukraine business https://www.theguardian.com/us-news/2024/feb/15/fbi-informant-biden-ukraine-charged-burisma
Ukraine:
Commander-in-Chief of the Armed Forces of Ukraine has officially: "Based on the operational situation around Avdiyivka, in order to avoid encirclement and preserve the lives and health of servicemen, I decided to withdraw our units from the city and move to defense on more favorable lines." https://x.com/CinC_AFU/status/1758642687417417845
“The 72nd mechanized brigade bogged down yet another Russian column, minus two BMP's and a T-62 tank.” https://x.com/NOELreports/status/1758514472635806055
“I have worked through the 10-page bilateral security commitment, which is now valid for 10 years between Ukraine and Germany (can be extended). The most important takes” https://x.com/deaidua/status/1758506704952521100
Sora is obviously not agent complete, nor would I describe its videos as scarily accurate... It's like people are seeing a totally different series of videos than I am and making wild conclusions. Surely we would have learned better by now...
The five scenarios you list -- where do they come from? I'm not sure if the Utopia scenario is a Utopia if we live on our own essentially.