Great work! And thanks for the lead     See Less

The self-justifying attitudes regarding climate change are alarming. Use government to hold their feet to the fire.     See Less

Here's a ChatGPT summary:

- The 2024 state of AI report was released, highlighting significant developments in AI.
- Models like Claude 3.5 Sonic, Grock 2, and Gemini 1.5 have     See More with GPT-4 due to overlapping pre-training data.
- OpenAI's projected 2024 costs for training a model are $3 billion, with expectations of not turning a profit until 2029.
- Meta's MovieGen model can produce audio and video simultaneously, showcasing impressive multimodal capabilities.
- Nobel Prizes in Physics and Chemistry were awarded to AI scientists, indicating AI's growing influence in scientific fields.
- Concerns about AI's future include potential threats compared to nuclear war and the possibility of AI becoming more intelligent than humans within 20 years.
- Brain LM, a model inspired by natural language models, can predict clinical variables from brain activity better than other methods.
- Chi1, an open-source alternative to AlphaFold 3, shows comparable or superior performance in certain domains.
- The release intervals of NVIDIA data center GPUs are decreasing, with increasing computational power, indicating rapid technological advancements.
- China is reportedly close to the AI frontier, despite restrictions on technology sales, through creative means of acquiring technology.
- The cost of AI models is dropping significantly, with an 80% price cut for equivalent performance in some cases.
- Copyright issues arise as companies like OpenAI and Runway ML scrape YouTube content for AI training, with new business models emerging to compensate creators.
- Environmental concerns are raised about AI's power usage, with some believing AI could eventually solve climate change.
- Jailbreaking AI models remains unsolved, with many techniques still effective, raising concerns about potential misuse.
- Predictions for 2025 include an open-source AI surpassing OpenAI's O1 and AI-generated research papers being accepted at major conferences, though these are debated.
- Main message: The AI landscape is rapidly evolving, with significant advancements, challenges, and ethical considerations shaping its future trajectory.    See Less

Jaron lanier makes a very persuasive argument in favour of ai systems having a kind of provenance record for the sources of information which were most important to them in learning about pa     See More ubjects and giving particular answers. Essentially the ai would be crediting high quality information sources. This would then make possible the paying of “royalties” to the creators of high quality information, which in turn would incentivise people to create more such high quality material, which would create a virtual circle of increasing quality of digital content. Lanier is well worth listening to on this aspect of ai. It doesn’t look as though anyone is heeding his advice, however!    See Less

Hello where are you ?     See Less

Wya mr ai explainedddddd     See Less

Streets ahead     See Less

It is 100% a wrong move to assume that AI will save us from climate change. Like Ilya said, we already know what we need to do. In fact, we already have most of the technological capabilitie     See More ically reduce carbon output, but the political will isn't quite there. Advancing the technology would help and I can see AI being part of that path, but I am extremely confident that ignoring climate goals now and instead putting all our bets on a superintelligence to retroactively fix everything for us is a terrible idea.    See Less

Worth noting: those "B200" (Blackwell) GPUs are only 2-3x faster at training, but reportedly they are over 10 times faster at inference.
The cost of test time compute is     See More rop dramatically.    See Less

'AI is eating the sciences' [3:39-5:10]     See Less