[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
Amy @_SFTahoe on x 19K followers
Created: 2025-07-19 02:08:44 UTC
IMO, Baidu is the best. It has the most compute, the best data and now the willingness to iterate to true vision only and AI. I am tracking several others as well, as well as US Startups. Many hope to rely on simulation instead of real world data. But it is the variations in real world data that help the neural nets recognize patterns and arrive at a generalizable solution—differences like dappled sunlight, glare, high angles of light at noon or low angles at high latitudes etc. Tesla neural nets learn from all these diverse scenarios for a robust solution.
Most are still relying on extensive sensors—as Elon Musk correctly noted sensors are a crutch to mask not smart enough neural nets due to a lack of data or compute or both. Same way high definition maps are a crutch. The issue is that neural nets training is not simple from video but is even more difficult from video; lidar point clouds etc. And sensor resolution adds a step and becomes problematic (can even cause a crash if bad data is believed or the extra sensor resolution caused latency delays a reaction too long. Aviation is littered with crashes from bad data. Like MCAS acting on bad data resulting in the Max crashes).
XXX engagements
