AI:
Google DeepMind announces AlphaGeometry: an AI system that solves Olympiad geometry problems at a level approaching a human gold-medalist. 📐
A neural language model (System 1, creative) with a symbolic deduction engine (System 2, reliable). It was trained solely on synthetic data and marks a breakthrough for AI in mathematical reasoning.
AlphaGeometry is a system made up of 2️⃣ parts:
🔵 A neural language model, which can predict useful geometry constructions to solve problems
🔵 A symbolic deduction engine, which uses logical rules to deduce conclusions
Both work together to find proofs for complex geometry theorems.
📊 In a benchmark test of 30 recent Olympiad geometry problems, AlphaGeometry solved 25 - surpassing the previous state-of-the-art approach called Wu’s method, which solved 10.
Notably, it solves all geometry problems in the 2000 & 2015 Olympiads, under human expert evaluation.
🔍 Unlike many prior methods, AlphaGeometry can produce human-readable proofs.
AlphaGeometry offers a potential framework to achieve advanced reasoning capabilities and discover new knowledge, without using human demonstrations.
This could help push forward theorem proving with AI - widely seen as a key step in building AGI. ↓
The code and model are open source: https://deepmind.google/discover/blog/alphageometry-an-olympiad-level-ai-system-for-geometry/
More AI links:
“OpenAI's next big model "will be able to do a lot, lot more" than the existing models can, CEO Sam Altman told Axios in an exclusive interview at Davos on Wednesday. Altman says AI advances will "help vastly accelerate the rate of scientific discovery." He doesn't expect that to happen in 2024, "but when it happens it's a big, big deal." Altman said his top priority right now is launching the new model, likely to be called GPT-5.” https://www.axios.com/2024/01/17/sam-altman-davos-ai-future-interview
Transformers are Multi-State RNNs: Shows that decoder-only transformers can be conceptualized as infinite multi-state RNNs—an RNN variant with unlimited hidden state size https://arxiv.org/abs/2401.06104
Intel presents Efficient LLM inference solution on Intel GPU https://arxiv.org/abs/2401.05391
A comprehensiveness study of trust and safety in LLMs. https://trustllmbenchmark.github.io/TrustLLM-Website/
A Unifying Framework for Inspecting Hidden Representations of Language Models https://arxiv.org/abs/2401.06102
Will ChatGPT soon get a memory? https://www.reddit.com/r/ChatGPTPro/comments/17wj0lp/chatgpt_is_getting_memory_soon/
Science and Technology:
Scientists Unveil Revolutionary SS-H2 Steel https://www.sciencedirect.com/science/article/abs/pii/S1369702123002390
Bladder tumours reduced by 90% using nanorobots https://ibecbarcelona.eu/bladder-tumours-reduced-by-90-using-nanorobots/
Miscellaneous:
“My most autistic quibble, which is correct, is that people who refuse to give probability estimates for things because "it is too uncertain" or gives "false precision" at a *deeply* fundamental level, misunderstand the purpose of probabilities.” https://twitter.com/shakoistsLog/status/1745156977070911629
On not being a radical medicine skeptic, and the dangers of doctor-by-Internet https://jakeseliger.com/2024/01/11/on-not-being-a-radical-medicine-skeptic-and-the-dangers-of-doctor-by-internet/
Politics:
Pakistan launches retaliatory strikes into Iran, with reports of seven killed https://www.bbc.com/news/world-asia-68014882
Ukraine:
“Amazing additional footage from the 🇷🇺Russian T-90M «Breakthrough» tank that got wrecked by the 25mm cannon of a 🇺🇦Ukrainian M2A2 Bradley IFV at Stepove, Donetsk Region.” https://twitter.com/GloOouD/status/1747741801451045078
Dmitry Medvedev: "There is a 100 percent probability of a new conflict, no matter what security papers the West signs with the puppet Kiev regime. It could happen in ten or fifty years," https://t.me/medvedev_telegram/437
"If the intensity of hostilities remains at the 2023 level, the Kremlin will be forced to take unpopular measures"-- regular deployment of units from other branches of armed forces and other security agencies, and a new wave of mobilization. https://ridl.io/the-russian-army-in-2024/
A Russian victory in Ukraine would create a world fundamentally antithetical to US interests and values with an empowered anti-Western coalition. https://www.understandingwar.org/backgrounder/high-price-losing-ukraine-part-2-%E2%80%94-military-threat-and-beyond
France's Minister of Defense, Sébastien Lecornu, emphasizes enduring military support for Ukraine amid Western aid fatigue. He reveals the production of 78 Caesar self propelled howitzers, a French artillery flagship, for Ukraine. https://www.leparisien.fr/international/sebastien-lecornu-il-faut-etre-endurant-dans-notre-soutien-militaire-a-lukraine-17-01-2024-CSY6LQXL2NCCXBKY7AMKKQTF7A.php
Russia has suffered a whole bunch of big defeats and failures, such as its inability to gain air superiority or destroy Ukraine's energy infrastructure. But its most embarrassing failure has to be its inability to stop Ukraine's Black Sea grain exports.
Ukrainian success in recent months: Black Sea grain exports are approaching pre-war levels.
(meant to write 2026!)
Ukraine 4 could have been strengthened with some more far-thinking "what now?" Yes, I agree Russia is a major threat but I am unsure whether that threat stops when the stalemate drags on into 2016? When the Russians are pushed back to the pre-2014 borders? Or when the people of Western Russia are freezing and starving in the dark after their infrastructure is crippled by long range strike weapons? How far do we have to be prepared to go?