Messages
4/15/2024
12 : 34
Import AI 369: Conscious machines are possible; AI agents; the varied uses of synthetic data
Who do you know who doesn't know about AI advances but should? ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
4/8/2024
12 : 4
Import AI 368: 500% faster local LLMs; 38X more efficient red teaming; AI21's Frankenmodel
If AI had truly superhuman capabilities, could we even effectively measure that? ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
4/1/2024
13 : 34
Import AI 367: Google's world-spanning model; breaking AI policy with evolution; $250k for alignment benchmarks
AI is neither rocket science nor brain surgery... which is very strange! ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
3/25/2024
12 : 44
Import AI 366: 500bn text tokens; Facebook vs Princeton; why small government types hate the Biden EO
plus: a story about a pleasant singularity ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
3/18/2024
12 : 44
Import AI 365: WMD benchmark; Amazon sees $1bn training runs; DeepMind gets closer to its game-playing dream
There is no absolution in being an absolutist. ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
3/11/2024
12 : 44
Import AI 364: Robot scaling laws; human-level LLM forecasting; and Claude 3
It increasingly feels like LLMs are observing us rather than the other way round. ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
3/4/2024
13 : 44
Import AI 363: ByteDance's 10k GPU training run; PPO vs REINFORCE; and generative everything
Things feel confusing right now because things are confusing right now - and they'll become more confusing. ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
3/4/2024
10 : 24
Import AI 360: Guessing emotions; drone targeting dataset; frameworks for AI alignment
Are there alternatives to the transformer which are roughly as compute efficient but entirely different in architecture?
3/4/2024
10 : 14
Import AI 361: GPT-4 hacking; theory of minds in LLMs; and scaling MoEs + RL
How many tokens would you need to represent your life? 10000? 100000? 1 million? ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
3/4/2024
9 : 23
Import AI 362: Amazon's big speech model; fractal hyperparameters; and Google's open models
The full capabilities of today's AI systems are all broadly undiscovered. ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
2/20/2023
16 : 14
Import AI 318: RL and addiction; Toolformer; and theology and AI.
AGI is a funhouse mirror of human ambition and most people define it in reference to a desired political outcome more than a technological outcome. View this email in your browser Welcome to Import AI,
2/6/2023
10 : 34
Import AI 317: DeepMind speeds up language model sampling; voice cloning tech gets abused; more scaling laws for RL
Right before a tsunami hits the tide goes way out and things seem to get oddly quiet. My general feel is that this is happening with some AI stuff where I have an inkling that labs are holding back or
1/30/2023
11 : 34
Import AI 316: Scaling laws for RL; Stable Diffusion for $160k; YOLOv8.
We are all agents. View this email in your browser Welcome to Import AI, a newsletter about artificial intelligence. Forward this email to give your chums an AI upgrade. Subscribe here. Here comes
1/20/2023
4 : 4
Import AI 315: Generative antibody design; RL's ImageNet moment; RL breaks Rocket League
I have vivid dreams these days about regular jobs that don't involve attempts to construct great machine minds. In these dreams I feel a great storm approaching from the horizon, full of
1/9/2023
12 : 34
Import AI 314: Language models + text-to-speech; emergent cooperation in wargames; ICML bans LLM-written papers
It's quite odd to be at the 'factional warfare' stage of AI development, but the ICML story shows how as AI progresses, we can expect the AI research community to be riven by ideological
1/3/2023
11 : 54
Import AI 313: Smarter robots via foundation models; Stanford trains a small best-in-class medical LM; Baidu builds a multilingual coding dataset
If AI timelines keep shortening among those working directly on AI systems, at what point do you need to 'hard update' policymakers that something strategic and important is closer than it
12/12/2022
12 : 24
Import AI 312: Amazon makes money via reinforcement learning; a 3-track Chinese AI competition; and how AI leads to fully personalized media
2022 will probably be the 'eternal september' of AI research; everyone is starting to pay attention View this email in your browser Welcome to Import AI, a newsletter about artificial
12/5/2022
11 : 34
Import AI 311: Distributed GPT busts the political economy of AI; Apple optimizes Stable Diffusion; AI war startup raises $1.48 billion
We live in an age of wonders, able to conjur up engines of creation that weave their own idiosyncratic synthesese from the threads of civilization. The West, after its dabble with a secular culture, is
11/28/2022
12 : 24
Import AI 310: AlphaZero learned Chess like humans learn Chess; capability emergence in language models; demoscene AI.
If I wasn't spending my dayjob working directly on AI and AI policy, I'd be running a scaled-up version of Import AI. Lately, I've been wondering if there's a way to do both - it'd
11/14/2022
14 : 4
Import AI 309: Generative bias; BLOOM isn't great; how China and Russia use AI
If we wanted to make the next five years of AI development go well, what would be the three most important things to work on, and what should be deprioritized? View this email in your browser Welcome