Niels Rogge@NielsRogge“So there’s this Chinese company called DeepSeek which basically does what OpenAI initially intended to do. They open-sourced a model trained with large-scale reinforcement learning, beating everyone else, and even releasing a paper detailing their process“Opens with an observation
Niels Rogge@NielsRoggeUnpopular opinion: benchmarks like these are moving the field in the wrong direction No I don't want an AI to be able to memorize (useless?) questions like "How many paired tendons are supported by a sesamoid bone?" in its weights I want the "intern", as @karpathy is suggestingOpens with a contrarian take
Niels Rogge@NielsRoggeSAM-3 is out on @huggingface! A big upgrade from SAM-2, and Meta finally added support for text prompts. Here I tried it out on @hazardeden10's magical goal against @Arsenal using the text prompt "Chelsea player" Works pretty well!Opens with an observation
Niels Rogge@NielsRoggeIt makes me a bit sad that amazing research like the one below isn't pursued a lot anymore due to the LLM era People just give up cause they think an LLM will beat them. Wrong. There are so many research directions to be explored, so many new architectures to be uncoveredOpens with an observation
Niels Rogge@NielsRoggeDon't let Anthropic fool you - it's literally just an LLM with scaled-up pre-training and post-training. As it's an LLM, it is only good at stuff humans have already done; it cannot invent new things. Anthropic themselves consider the catastrophic risks lowOpens with a "stop doing this"